265SmithWatt 75Neumann JHuangDHassabisFLiEMusk 20 Agentic AIforU

KingCharlesLLM DeepLearning009 NormanMacrae.net EconomistDiary.com Abedmooc.com

Update Taiwan AI june 2026- Nvidia 150 company ecosystem in Taiwan- 1000 agent CUDA-X platforms - ;;V10+8* (Vera Rubin 10) replaces G10 in strategic impact as Intelligence Nation Builders -example Japan role in V10; RTX spark transforms PC market; plan B if education systems continue to fail Jensen's Test, Cosmos Robot platform  celebrating AI as greatest job creating leap ..jensen computex +14 pre-game scenesetters.  Mooc for parents Celebrating Jensen Test

Huang Observing AI's biggest deals - 2012     1984    2026 & moments 1956 1905

https://www.youtube.com/watch?v=b1EaoN2CnCE Join Jensen live from Taiwan May 31 East Coast USA -pregame 9pm , jensen 11pm .. and meet Robots

June 1  in Taiwan -- pretrain with JENSEN's TEST

PREVIOUSLY

May 7 20000 DC brains thank SCSP AI+Expo. May 13 Trump asks Huang to join rest of China entourage at Alaska

Jensen video with stanford https://www.youtube.com/watch?v=tsQB0n0YV3k selected quotes

ED: Breaking 12 May 2026 AI Best kept secret- its all too much talk unless layer 5 AI connects massive application community scaling 

.
GEMINI review 5/17/2026:: Governments and national champions in these regions are explicitly aligning their "Sovereign AI" policies with this layered approach to ensure they own their intelligence production.
Asia-Pacific Taiwan: Positioning itself as an "AI Island." It leverages its chip dominance (Layer 2) while investing over $3 billion in energy and "AI factories" (Layers 1 & 3). Singapore: Launched the first Model AI Governance Framework for Agentic AI in 2026 and is building a "Southeast Asia AI Plan" to localize the stack.Japan & South Korea: Rapidly building domestic "AI factories" to reduce dependence on foreign cloud providers.India: Implementing a Sovereign AI Plan that includes state-level LLM development and massive infrastructure buildouts. Indonesia & Vietnam: Emerging hubs for data center (Layer 3) and energy (Layer 1) partnerships to support regional AI growth
. 

Middle East  United Arab Emirates = Early adopters of the "Sovereign AI" narrative. The UAE’s AI Minister and NVIDIA have collaborated to frame AI as a "national natural resource". Saudi has made big purchases of Vera Rubin directly and with AWS

Europe France: Home to Mistral, which partnered with NVIDIA for a 18,000-chip data center to create a European "homegrown alternative". Germany & Netherlands: Focus on high-end manufacturing (ASML) and industrial AI applications (Layer 5). Nordics: Using their green energy surplus (Layer 1) to attract the world's most sustainable "AI Factories". UK: Prioritizing "compute sovereignty" through large-scale government-funded supercomputing projects

Gemini map review 5/17/2026continued here

May 5-9 Washington DC leaps closer to Mapping Huang AI. Even By Jensen Huang norms its been a massive weekend for democratising AI. Looks like 20000 turned up to Carnegie Melon commencement address Sunday May 10 , by Saturday 20000 DC brains had been greeted by layer 5 AI community apps: AI+expo www.scsp.ai

AI REVIEWS 10 MOST IMPOERANT AI (layer 5) APPLIUCATIONS FOR  STUDENTS & TEACHERS OF 2025-6

Do any of top 10 surprise you? Some Westerners are surprised to see Robotics and Embodied AI at number 3. WE expect Asian supercities will lead with introduction not juts of self driving cars (almost ready after 15 years of AI research) but eg with public services in Tokyo where the elderly population dynamics cant find enough people to maintain the city world class usability standards - try benchmarking Bloomberg top 10 ai cities; The two academics with the deepest data algorithm fei-fei li and yann lecun have moved on to world ai and embodied ai. There's not much point in AI having the greatest brainpower in the world if a 2 year toddler understands gravity's consequences more actionably.  Turns out its not just kids who need to celebrate Head, Shoulders, Knees & Toes (and arms); each requiring practice....

But where did vision that ai needs youth to scale its communal applications begin? As far as we can see one bleak December day CSIS Dec 2025 - please help us make sure joyous Diary of layer 5 AI is complete https://normanmacrae.ning.com/forum/topics/jensen100 Hopefully next school year AI curricula wont exist without it : JH CSIS December; India Ambassador DC CSIS Jan ; French & India Ambassador Brookings; India AI Summit Delhi; 2000 AI Expo www.scsp.ai 20000+ carnegie melon commencement

===================================

Breaking - help us video tour every placesmost youth inspiring intelligences

Breaking 16 March2026 at nvidia youtube today replay Jensen GTU speech - his global update of partners and purposes of computers with billion times more maths brainpower than separate human minds- note search out last time jensen keynoted in place you care about for more personalised ai agency nb jensen cases cover all tech needs to bring together to advance human work - eg 5 layers of full stack ai - energy , supercomputer chips, place unique investment purpose of tech and peoples, ai models, scaling community actions/aps. There are at least 3 gameboards beyond LLM agency: Einsteins' solutions to natures biggest puzzles such as illnesses and resource sustainability, space, how world class citizens demand last mile services from autonomous robots and eg drones.

AGENCY & OPEN CLAW GET 2026  OFF TO IMAGINATION GREATEST YEAR?

Breaking Jensen Huang hosts open model session

Open Models: Where We Are and Where We’re Headed [S82480] March 18 3.30est

Harrison Chase | Co-Founder and CEO | LangChain

Jensen Huang | Founder and CEO | NVIDIA

Michael Truell | CEO and Co-Founder | Cursor

Misha Laskin | Co-founder and CEO | Reflection AI

Aravind Srinivas | CEO & Co-founder | Perplexity

Mira Murati | Founder and CEO | Thinking Machines Lab

Arthur Mensch | Co-Founder and CEO | Mistral

Daniel Nadler | CEO | OpenEvidence

Hanna Hajishirzi | Professor, Sr. Director NLP | AI2

Anjney Midha | Founder | Founder, AMP PBC

Robin Rombach | Co-Founder and CEO | Black Forest Labs

IN CHINA ITS THE YEAR OF THE LOBSTER (launching 300 AI team)

Help us clarify which ai superstars you may want to know Case Jensen Huang One of keys to life in 2026 is understand 3 million fold tech changes: what did people do with million fold more chip power 1965-95 then 95-2025 what did they do with million times deeper computer brainpower acceleration and million fold more data transmitted by satellites between communities and clouds. Jensen Huang connected more partnerships than anyone else we know in deeper compute
P1.1 // P2.1 \\ Technology is useless unless it benefits everybody. Jensen Huang has done most to design machines with billion times more maths brainpower than separate human minds. He's up for helping anywhere that wants to democratise intelligence eg  2026India Feb  indiasovai26-1.docx  , Santa Clara March, DC May SCSP.ai ai+expo
To non-mathematicians Jensen says dont worry; discuss purposes of peoplewho bring you AI. Connect your peer network(or home community) to someone transparently scaling the purpose your peoples most need to advance 
Help clarify purposes Jensen is helpig diggerent places with - which jensen keynote are you interested in rsvp chris.macrae@yahoo.co.uk - example washington dc october 205 jensen updated 5 purposes:
TECH MANUFACTURING: His Taiwan partners TSMG & Foxconn have brought trump factories manufacturing superchips and advanced consumer tech
AI FACTORY Nvidia partners have helped us national supercomputers run by Department of Energy install 7/8 AI factory supercomputers - lead applications include biotech, energy ,
US QUANTUM He has arranged for about 15 Quantum boutiques to help advance US quantum supporting energy and other deep science need   
US 6G he has arranged for Nvidia to help US regain 6g leadership
US Last mile Transportation Autonomy With uber he's arranged for US to keep track of where world cllass cities are benchmarking both self driving cars and any last mile advances in autonomous machines
NB Nvidia Jensen is unique position to help with purposes because he has not only built the supercomputers AI performs on but cuda application libraries of most of the world leading uses of AI
.
 
click to GROK Answer AO,  to Q0 at ASAI - question 0 if there are at least 4 different jensen AI wonderlands, why is so much chat distracted by one AGI?

Apps-Agents

Million AI Mods

Data Sov Invest  Platforms

Chips

Energy

........

........
help build llm fed top 100 jensen huang actions -LInkedin:  chris.macrae@yahoo.co.uk
. .
................................
ALPAMAYO: happy ai 2026 wonderful news https://www.youtube.com/live/L0jJrzFMbP8?si=RNIy98UOclj9c2hZ&t=... mercedez benz has designed car with 25+ sensors which is not only driving's safest but could be intelligence data any city needs to transform from being run by gangs 4 best chances to understand this ces LV yesterday, nvidia santa clara march, scsp.ai ai+expo dc may, computex taiwan june?
help needed - updating list of jensen huang INTELLIGENCE  top 100 videos and transcripts fed into llm open micro agent JH -nb some dates to verify (my viewing versus first posted)
Dec 27, 2025 from nox queen elizabeth engineering awards
-see also king Charles hands Huang copy of personal questions 2024-2025
(3447) Oral History of Jensen Huang - YouTube www.youtube.com
Dec 25, 2025 
(3447) Nvidia Acquires Groq for $20 Billion: Why Jensen Huang Just Bought the Future - YouTube www.youtube.com
Dec 25, 2025 

(3397) SCSP.ai  Episode 31: A Conversation with Jensen Huang - YouTube www.youtube.com
Dec 19, 2025 
Jensen Huang Is NOT Backing Down — Google & AWS Bring Their Hardest AI Chips Yet - YouTube www.youtube.com
Dec 14, 2025 
(3346) NVIDIA’s Jensen Huang on Securing American Leadership on AI - YouTube www.youtube.com
Dec 14, 2025 
(3342) Jensen Huang: AI Energy War, US–China Chip Rivalry and Next Revolution|黃仁勳最新訪談:AI能源戰、美中晶片攻防與下一場工業革命 - YouTube www.youtube.com
Dec 13, 2025 

Dec 6, 2025
jensen Stuns Washington DC thinktank clarifies 5 layer AI model : 1 energy; 2 chipsd ghardware; 3 software platforms applied as data sovereignty investments
4 1.5 million ai models
5 billions  of action aps at edge mainly generated by open ai models

LIVE | Macron Tours VivaTech 2025, Joins Nvidia CEO Jensen Huang & Mistral's Arthur Mensch | APT - YouTube www.youtube.com 
Dec 6, 2025 reprise of summer breakthrough - macron sys jensen has done his job for him - can now retire with france and eu intelligence opening up

(3330) How Jensen Huang’s Children Quietly Took Over Nvidia - YouTube www.youtube.com
Dec 4, 2025 -well actually this is about nvidia future of humanoids serving safe cities - it also raises question - who's collecting last mile data (surprsie at google eats and aurora- brookings 555penn - mid december

(3240) A Conversation with NVIDIA CEO Jensen Huang – The Future of AI and Energy - YouTube www.youtube.com
Dec 1, 2025

A Conversation with NVIDIA CEO Jensen Huang – The Future of AI and Energy - YouTube www.youtube.com
Dec 1, 2025

(3357) "jensen huang" - YouTube www.youtube.com
Dec 14, 2025

(3346) NVIDIA’s Jensen Huang on Securing American Leadership on AI - YouTube www.youtube.com
Dec 14, 2025
(3342) Jensen Huang: AI Energy War, US–China Chip Rivalry and Next Revolution|黃仁勳最新訪談:AI能源戰、美中晶片攻防與下一場工業革命 - YouTube www.youtube.com

Dec 13, 2025
(3346) NVIDIA’s Jensen Huang on Securing American Leadership on AI - YouTube www.youtube.com

Dec 12, 2025
(3330) How Jensen Huang’s Children Quietly Took Over Nvidia - YouTube www.youtube.com

Dec 11, 2025
(3322) NVIDIA’s Jensen Huang on Securing American Leadership on AI - YouTube www.youtube.com

Dec 11, 2025
(3346) NVIDIA’s Jensen Huang on Securing American Leadership on AI - YouTube www.youtube.com

Dec 11, 2025
NVIDIA’s Jensen Huang on Securing American Leadership on AI - YouTube www.youtube.com

Dec 11, 2025
(3346) NVIDIA’s Jensen Huang on Securing American Leadership on AI - YouTube www.youtube.com

Dec 9, 2025
NVIDIA’s Jensen Huang on Securing American Leadership on AI - YouTube www.youtube.com

Dec 9, 2025
(3307) Jensen Huang’s Keynote Highlights at NVIDIA GTC Washington, D.C. - YouTube www.youtube.com

Dec 9, 2025
(3307) NVIDIA CEO Jensen Huang Warns: China May Outperform the U.S. in Tech | Full Interview | AI1G - YouTube www.youtube.com

Dec 9, 2025
(3303) NVIDIA’s Jensen Huang on Securing American Leadership on AI - YouTube www.youtube.com

(2963) Nvidia's Jensen Huang on AI & the Next Frontier of Growth - YouTube www.youtube.com
Oct 27, 2025

NVIDIA's Founder & CEO, Jensen Huang Fireside Chat With Mukesh Ambani | NVIDIA AI Summit | CNBC TV18 - YouTube www.youtube.com
Oct 24, 2025

(2943) Nvidia's Jensen Huang Hand-Delivers World's Smallest Supercomputer To Elon Musk - YouTube www.youtube.com
Oct 23, 2025

(2963) Nvidia's Jensen Huang on AI & the Next Frontier of Growth - YouTube www.youtube.com
Oct 23, 2025

(2943) jensen huang - YouTube www.youtube.com
Oct 23, 2025

(2943) jensen huang - YouTube www.youtube.com
Oct 23, 2025

(2943) How Jensen Works - YouTube www.youtube.com
Oct 23, 2025

Found 118 search results for 'youtube jensen' (2942) Jensen Huang Special Address from NVIDIA AI Summit Japan - YouTube www.youtube.com
Oct 23, 2025
APEC CEO Summit - NVIDIA JHH Special Address
November 2025
, NVIDIA
01:43:08
NVIDIA GTC Washington, D.C. Keynote
NVIDIA GTC Washington, D.C. Keynote
October 2025
, Founder and CEO, NVIDIA
01:36:55
GTC Paris Keynote at VivaTech
GTC Paris Keynote at VivaTech
June 2025
, Founder and CEO, NVIDIA
01:42:51
NVIDIA CEO Jensen Huang Keynote at COMPUTEX 2025
NVIDIA CEO Jensen Huang Keynote at COMPUTEX 2025
May 2025
, Founder and CEO, NVIDIA
02:11:48
GTC 2025 Keynote
GTC 2025 Keynote
March 2025
, Founder and CEO, NVIDIA
02:03:20
Quantum Computing: Where We Are and Where We’re Headed
Quantum Computing: Where We Are and Where We’re 
March 2025
, Founder and CEO, Atom Computing
, CEO, D-Wave
, Technical Fellow, Microsoft
, CEO and Co-Founder, SEEQC
, Founder and CEO, NVIDIA
, Chief Executive Officer , Pasqal
, CEO, Infleqtion
, Joshua and Beth Friedman University Professor, QuEra Computing
, General Manager, Quantum Technologies, AWS
, CEO, Rigetti
, Co-Founder and Chief Scientific Officer, PsiQuantum
, Executive Chair, IonQ
, Chief Scientist and Co-Founder, Quantum Circuits
, President and CEO , Quantinuum
, CEO and Co-Founder , Alice & Bob
01:36:59
NVIDIA CEO Jensen Huang Keynote at CES 2025
NVIDIA CEO Jensen Huang Keynote at CES 2025
February 2025
, Co-founder, president and CEO, NVIDIA
NVIDIA SC24 Special Address
NVIDIA SC24 Special Address
January 2025
, VP & GM, Hyperscale & HPC, NVIDIA
, NVIDIA
01:26:51
AI Summit Japan 2024 Fireside Chat
AI Summit Japan 2024 Fireside Chat
November 2024
, Chairman and CEO, Softbank
, Founder, President and CEO, NVIDIA Corporation
44:38
NVIDIA CEO Jensen Huang’s Special Address at AI Summit India
NVIDIA CEO Jensen Huang’s Special Address at AI 
October 2024
, Founder and CEO, NVIDIA
58:36
AI and The Next Computing Platforms With Jensen Huang and Mark Zuckerberg
AI and The Next Computing Platforms With Jensen 
August 2024
, Founder and CEO, Meta
, NVIDIA
59:25
What’s Next in AI: NVIDIA’s Jensen Huang Talks With WIRED’s Lauren Goode
What’s Next in AI: NVIDIA’s Jensen Huang Talks With 
August 2024
, Sr. Writer, WIRED
, NVIDIA
NVIDIA CEO Jensen Huang Keynote at COMPUTEX 2024
June 2024
, Founder and CEO, NVIDIA
01:00:35
Replay – GTC Korean AI Expert Day
Replay – GTC Korean AI Expert Day
May 2024
, NVIDIA
, NVIDIA
, NVIDIA
53:47
Transforming AI
Transforming AI
March 2024
, Co-Founder and Chief Executive Officer, Cohere
, Co-Founder and CEO, Essential AI
, Co-Founder and Chief Executive Officer, Inceptive
, Founder and CEO, NVIDIA
, Co-Founder, NEAR Protocol
, CTO, Sakana AI
, Member of Technical Staff, OpenAI
, Chief Executive Officer and Co-Founder, Character.AI
49:08
Earth-2: Updates on kilometer-scale visualization, simulation, digital twinning, and AI super-resolution
Earth-2: Updates on kilometer-scale visualization,
March 2024
, Principal Engineer and Scientist, AI-HPC , NVIDIA
, Dir. of Climate Simulation Research , NVIDIA
02:03:08
GTC 2024 Keynote
GTC 2024 Keynote
March 2024
, Founder and CEO, NVIDIA
04:54
NVIDIA Nsight Tools Ecosystem
NVIDIA Nsight Tools Ecosystem
February 2024
, Product Marketing Manager, NVIDIA
12:21
NVIDIA CEO Jensen Huang at Microsoft Ignite 2023
NVIDIA CEO Jensen Huang at Microsoft Ignite 2023
January 2024
, CEO, Microsoft
, NVIDIA
07:37
Jensen Huang at VMware Explore 2023 - General Session Keynote
Jensen Huang at VMware Explore 2023 - General 
August 2023
, CEO, VMWare
, Founder and CEO, NVIDIA
02:30
VMware Private AI Foundation with NVIDIA – Readying Enterprises for Generative AI
VMware Private AI Foundation with NVIDIA – Readying 
August 2023
, CEO, VMWare
, Founder and CEO, NVIDIA
01:22:03
NVIDIA Keynote at SIGGRAPH 2023
NVIDIA Keynote at SIGGRAPH 2023
August 2023
, NVIDIA
01:55:57
NVIDIA Keynote at COMPUTEX 2023
NVIDIA Keynote at COMPUTEX 2023
June 2023
, Founder and CEO, NVIDIA
01:05:31
Level Up with NVIDIA: Nsight Graphics - How to Optimize your Game
Level Up with NVIDIA: Nsight Graphics - How to Optimize 
April 2023
Fireside Chat with Ilya Sutskever and Jensen Huang
March 2023
, Founder and CEO, NVIDIA
, Co-founder and Chief Scientist, OpenAI
01:17:51
GTC 2023 Keynote
GTC 2023 Keynote
March 2023
, Founder and CEO, NVIDIA

GTC

14 VIDEOS (23h 28m)
  • GTC 2023 Keynote with NVIDIA CEO Jensen Huang

    GTC 2023 Keynote with NVIDIA CEO Jensen Huang
    MARCH 19, 2023
  • GTC Sept 2022 Keynote with NVIDIA CEO Jensen Huang

    GTC Sept 2022 Keynote with NVIDIA CEO Jensen Huang
    SEPTEMBER 18, 2022
  • GTC 2022 Keynote with NVIDIA CEO Jensen Huang

    GTC 2022 Keynote with NVIDIA CEO Jensen Huang
    MARCH 20, 2022
  • GTC Nov 2021 Keynote with NVIDIA CEO Jensen Huang

    GTC Nov 2021 Keynote with NVIDIA CEO Jensen Huang
    NOVEMBER 7, 2021
  • GTC 2021 Keynote with NVIDIA CEO Jensen Huang

    GTC 2021 Keynote with NVIDIA CEO Jensen Huang
    APRIL 10, 2021
  • GTC 2019 Keynote with NVIDIA CEO Jensen Huang

    GTC 2019 Keynote with NVIDIA CEO Jensen Huang
    MARCH 19, 2019
  • GTC 2018 Keynote with NVIDIA CEO Jensen Huang

    GTC 2018 Keynote with NVIDIA CEO Jensen Huang
    APRIL 9, 2018
  • GTC 2017 Keynote with NVIDIA CEO Jensen Huang

    GTC 2017 Keynote with NVIDIA CEO Jensen Huang
    MAY 10, 2017
  • GTC 2016 Keynote with Gill Pratt, CEO of Toyota Researc

    GTC 2016 Keynote with Gill Pratt, CEO of Toyota Researc
    APRIL 7, 2016
  • GTC 2016 Keynote with Rob High, CTO of IBM Watson

    GTC 2016 Keynote with Rob High, CTO of IBM Watson
    APRIL 6, 2016
  • GTC 2016 Opening Keynote with NVIDIA CEO Jensen Huang

    GTC 2016 Opening Keynote with NVIDIA CEO Jensen Huang
    APRIL 5, 2016
  • GTC 2015 Keynote with Dr. Andrew Ng, Baidu

    GTC 2015 Keynote with Dr. Andrew Ng, Baidu
    MARCH 19, 2015
  • GTC 2015 Keynote with Jeff Dean, Google

    GTC 2015 Keynote with Jeff Dean, Google
    MARCH 18, 2015
  • GTC 2015 Opening Keynote with NVIDIA CEO Jensen Huang

    GTC 2015 Opening Keynote with NVIDIA CEO Jensen Huang
    MARCH 17, 2015

Views: 778

Reply to This

Replies to This Discussion

________________ [VIDEO NARRATION]   Cadence and NVIDIA are partnering to build chip  design agents. Hundreds of thousands of NVIDIA  

1:12:53

1 hour, 12 minutes, 53 seconds

chips come together to make the AI factories  that power the world’s frontier AI models.  

1:12:59

1 hour, 12 minutes, 59 seconds

Designing these chips, and the systems they run  in, is one of the hardest engineering challenges.  

1:13:05

1 hour, 13 minutes, 5 seconds

Trillions of transistors. Three-dimensional  circuits at microscopic scale. Every gate,  

1:13:11

1 hour, 13 minutes, 11 seconds

every wire, synchronized to picoseconds, must work  in perfect harmony, with no margin for error.  

1:13:18

1 hour, 13 minutes, 18 seconds

Physical prototypes are too slow and too costly.  So engineers work in the digital realm. Each chip  

1:13:24

1 hour, 13 minutes, 24 seconds

begins as a set of architectural specifications,  then translated into RTL — the language of chip  

1:13:30

1 hour, 13 minutes, 30 seconds

design. RTL must be verified in simulation. A  single bug can delay a chip by months. At NVIDIA,  

1:13:38

1 hour, 13 minutes, 38 seconds

thousands of engineers, billions of compute  hours per year, millions of tests written, run,  

1:13:44

1 hour, 13 minutes, 44 seconds

and debugged — a cycle that takes teams weeks. To compress this cycle, Cadence and NVIDIA built  

1:13:50

1 hour, 13 minutes, 50 seconds

a design verification agent. Codex orchestrates  the process. Cadence Chip Stack launches the  

1:13:57

1 hour, 13 minutes, 57 seconds

RTL verification loop, powered by Nemotron  and secured by NVIDIA OpenShell. Calling on  

1:14:03

1 hour, 14 minutes, 3 seconds

expert subagents in RTL generation, test bench  creation, regression testing, and debug. The  

1:14:11

1 hour, 14 minutes, 11 seconds

system drives itself. The Chip Stack agents run  hundreds of simulations with Cadence Xcelium,  

1:14:17

1 hour, 14 minutes, 17 seconds

formal verification with JasperGold. Design flaws  revealed. Bugs in the code, fixed. What once took  

1:14:24

1 hour, 14 minutes, 24 seconds

weeks now takes hours. Verification cycles over  40 times faster. Together, NVIDIA and Cadence are  

1:14:32

1 hour, 14 minutes, 32 seconds

reinventing chip design with AI agents. ________________   [JENSEN] From weeks to hours.  

1:14:41

1 hour, 14 minutes, 41 seconds

From weeks to hours. From weeks to hours. NVIDIA has thousands of chip designers.  

1:14:47

1 hour, 14 minutes, 47 seconds

We are going to hire hundreds of thousands  of Cadence super agents to work with us,  

1:14:54

1 hour, 14 minutes, 54 seconds

so that we can accelerate our company,  so that we can be even more ambitious,  

1:14:59

1 hour, 14 minutes, 59 seconds

create even more amazing things, run even faster. You saw earlier that the toolkit — with models,  

1:15:08

1 hour, 15 minutes, 8 seconds

harness, tools and skills, and runtimes — is  the reason why we’re working with Cadence so  

1:15:18

1 hour, 15 minutes, 18 seconds

hard to accelerate all of their tools on CUDA,  because the agents are impatient. The agents  

1:15:25

1 hour, 15 minutes, 25 seconds

want the answer immediately. And so, models,  harnesses, CUDA-accelerated libraries and tools,  

1:15:36

1 hour, 15 minutes, 36 seconds

and then the runtime. What you saw just  now is all of that coming together.  

1:15:41

1 hour, 15 minutes, 41 seconds

Now, one of the things that it starts with  is a great model that Cadence could modify  

1:15:48

1 hour, 15 minutes, 48 seconds

and tune to be expert at the Cadence  workflow, at the Cadence expertise,  

1:15:53

1 hour, 15 minutes, 53 seconds

so that they could create super agents that are  proprietary to Cadence, with their proprietary  

1:16:00

1 hour, 16 minutes

knowledge. They have to start with an excellent  model. We call it Nemotron. It is dedicated to  

1:16:07

1 hour, 16 minutes, 7 seconds

building open models for the world, so that all  of you, all of us, could create our own agents.  

1:16:14

1 hour, 16 minutes, 14 seconds

Today, we’re announcing Nemotron 3 Ultra. Our  next open model, and it is smart. The Nemotron  

1:16:32

1 hour, 16 minutes, 32 seconds

models not only give you the model — we give  you all the data that we use to train the model,  

1:16:40

1 hour, 16 minutes, 40 seconds

and because we have a coalition of incredible  partners, we work together and contribute data  

1:16:49

1 hour, 16 minutes, 49 seconds

to each other. Nemotron is trained on one of the  largest suites of long-running reasoning models,  

1:16:56

1 hour, 16 minutes, 56 seconds

long-running tool task-solving, tool-using  datasets in the world, because of all of our great  

1:17:04

1 hour, 17 minutes, 4 seconds

partnerships. All of this — from the model, the  training scripts, and the data — made completely  

1:17:12

1 hour, 17 minutes, 12 seconds

available to you. This is open models at its  best. The best open model system in the world.  

1:17:20

1 hour, 17 minutes, 20 seconds

Simple goal: so that you can take all of it,  add to it, make it even better, make it yours.  

1:17:27

1 hour, 17 minutes, 27 seconds

Nemotron 3 Ultra is five times faster. This is the  world’s first model based on a hybrid architecture  

1:17:37

1 hour, 17 minutes, 37 seconds

of SSM — State Space Models — with a Mixture of  Experts. The architecture is incredibly fast.  

1:17:46

1 hour, 17 minutes, 46 seconds

We made it fast so that you could think fast. When  you think fast, you can think longer at the same  

1:17:50

1 hour, 17 minutes, 50 seconds

cost. So five times faster. It is also 30% cheaper  — 30% lower cost to run in total FLOPs and total  

1:18:02

1 hour, 18 minutes, 2 seconds

inference time — than even the most cost-effective  models in the world. We’re comparing against the  

1:18:07

1 hour, 18 minutes, 7 seconds

world’s best open models. Frontier smart. Five  times faster, 30% cheaper. Completely open. We’re  

1:18:19

1 hour, 18 minutes, 19 seconds

completely dedicated to this. This is now Nemotron  3. We’re currently working on Nemotron 4.  

1:18:26

1 hour, 18 minutes, 26 seconds

So this entire toolkit — from models, harnesses,  tools and skills, and runtimes — is the reason  

1:18:35

1 hour, 18 minutes, 35 seconds

why every enterprise company in the world has  the ability now to create their own agents,  

1:18:43

1 hour, 18 minutes, 43 seconds

just like Cadence did with their super agents.  And we’re working with so many companies — Cadence  

1:18:48

1 hour, 18 minutes, 48 seconds

and CrowdStrike, ServiceNow and Palantir, SAP.  People always said, “Jensen, the agents are going  

1:18:56

1 hour, 18 minutes, 56 seconds

to disrupt these markets.” I said: completely the  opposite. And you can now see it. Agents are going  

1:19:03

1 hour, 19 minutes, 3 seconds

to create the largest opportunity ever for my  partners and friends. And we have the NVIDIA Agent  

1:19:12

1 hour, 19 minutes, 12 seconds

Toolkit for Enterprise AI to help them. So there you go.  

1:19:23

1 hour, 19 minutes, 23 seconds

First, Vera Rubin in full production. Two,  Vera CPU — CPU built for a new generation,  

1:19:31

1 hour, 19 minutes, 31 seconds

for agents. And three, NVIDIA’s  Enterprise AI Toolkit, so that  

1:19:38

1 hour, 19 minutes, 38 seconds

every enterprise and every enterprise  software company can build agents.  

1:19:53

1 hour, 19 minutes, 53 seconds

My relationship with you started here.  And many of you, many of my friends and  

1:20:01

1 hour, 20 minutes, 1 second

partners here in Taiwan — your companies  started here. This is, in a lot of ways,  

1:20:09

1 hour, 20 minutes, 9 seconds

the beginning of the modern computer industry  — 40 years now. And NVIDIA is 33 years old.  

1:20:16

1 hour, 20 minutes, 16 seconds

The PC industry was already starting to get going  — Windows 1, Windows 2, Apple 1, Apple 2 — and by  

1:20:26

1 hour, 20 minutes, 26 seconds

the time we came along, Windows 3.1 was the PC.  And as you know, Windows 95 made the PC personal.  

1:20:36

1 hour, 20 minutes, 36 seconds

It took the PC from enterprises and companies  and made it into a consumer electronics device.  

1:20:43

1 hour, 20 minutes, 43 seconds

Everybody should have one, and everybody does. This is the beginning. This computing platform  

1:20:49

1 hour, 20 minutes, 49 seconds

did several things incredibly smart.  Windows was not just disaggregated,  

1:20:55

1 hour, 20 minutes, 55 seconds

as you know. Windows was properly abstracted.  It was architected just right — system BIOS,  

1:21:03

1 hour, 21 minutes, 3 seconds

open chipsets, the operating system with drivers  that could be connected and installed at runtime,  

1:21:15

1 hour, 21 minutes, 15 seconds

and an abstraction layer with a multimedia  API that opened up the PC to what we all  

1:21:24

1 hour, 21 minutes, 24 seconds

know today. Each one of these elements was  essential in making the PC so popular.  

1:21:32

1 hour, 21 minutes, 32 seconds

40 years later, Microsoft and NVIDIA are going  to reinvent the PC. This is going to be the new  

1:21:41

1 hour, 21 minutes, 41 seconds

  1. Now, tomorrow night I’m going to be with  Satya. We’re going to talk a lot more about  

1:21:50

1 hour, 21 minutes, 50 seconds

the work that we’re doing together. Microsoft and  NVIDIA — over the last three years, it took this  

1:21:57

1 hour, 21 minutes, 57 seconds

long to completely reinvent how the PC is going to  work so that we could be ready for this moment.  

1:22:04

1 hour, 22 minutes, 4 seconds

.........................................................................................................................................................................................................................................................................................................................RTX Spark -changes pc market for first time in 60 years - see supplier ecosystem and partners accelerating AI PC world- scroll to 1.30 40  of Jensen Taiwan Speech 1 June 2026 _related Jensen Test education (job-positive) transformation

1:22:04

1 hour, 22 minutes, 4 seconds

As I mentioned earlier, the computing pattern  called an agent is going to run in AI clouds.  

1:22:11

1 hour, 22 minutes, 11 seconds

It’s going to run inside enterprises. It is also  going to run on your PC. What’s going to happen  

1:22:19

1 hour, 22 minutes, 19 seconds

to that PC when it has an autonomous agent? An  agent that’s helping you, that understands you.  

1:22:26

1 hour, 22 minutes, 26 seconds

You could talk to it. It could look at you.  You can ask it to refile things, go help you,  

1:22:33

1 hour, 22 minutes, 33 seconds

do some research. It could do a lot more than  I’ll show you. But the new operating system is,  

1:22:40

1 hour, 22 minutes, 40 seconds

of course, the old operating system plus large  language models. Large language models in a lot  

1:22:47

1 hour, 22 minutes, 47 seconds

of ways are the modern version of DirectX. It has,  of course, input and output, understands prompts,  

1:22:55

1 hour, 22 minutes, 55 seconds

it understands computer vision, it can generate  video, it can generate sound. It is the modern  

1:23:00

1 hour, 23 minutes

extension — the intelligence extension —  of the PC. On top of that, the application,  

1:23:09

1 hour, 23 minutes, 9 seconds

as I mentioned before, is going to be replaced  by an agentic runtime. And that is the modern  

1:23:16

1 hour, 23 minutes, 16 seconds

application: an agent. Let’s take a look. ________________   [VIDEO NARRATION] It started with a spark. An  

1:23:26

1 hour, 23 minutes, 26 seconds

idea — to reimagine the PC for the first time in  40 years, for the age of AI. What becomes of our  

1:23:36

1 hour, 23 minutes, 36 seconds

personal computer in a world of agents? Agents  running natively, connected to models — local  

1:23:43

1 hour, 23 minutes, 43 seconds

or in the cloud — are personal AIs, sandboxed  for security, running continuously, getting  

1:23:51

1 hour, 23 minutes, 51 seconds

work done. The chips and the OS must evolve. Introducing RTX Spark — everything we’ve learned  

1:24:01

1 hour, 24 minutes, 1 second

over 33 years, distilled into one chip. Blackwell  RTX GPU with 6,144 Tensor Cores. One petaflop of  

1:24:12

1 hour, 24 minutes, 12 seconds

AI performance. A custom 20-core Grace CPU, built  in partnership with MediaTek. Fused by NVLink. 128  

1:24:24

1 hour, 24 minutes, 24 seconds

gigabytes of unified memory. TSMC 3-nanometer  process. 70 billion transistors. And in close  

1:24:34

1 hour, 24 minutes, 34 seconds

collaboration with Microsoft — a Windows platform  for agents. We’re reinventing the personal  

1:24:42

1 hour, 24 minutes, 42 seconds

computer. We’re creating for the age of agents.  This is the dawn of a new personal computing  

1:24:54

1 hour, 24 minutes, 54 seconds

revolution. And it starts with NVIDIA RTX Spark. ________________  

1:25:04

1 hour, 25 minutes, 4 seconds

[JENSEN] Here it is. Of course,  

1:25:18

1 hour, 25 minutes, 18 seconds

I’ve got to show you the most beautiful  part — which is video games. It’s also  

1:25:23

1 hour, 25 minutes, 23 seconds

the closest to our heart. This is Forza.  This is 007, by the way — the new 007 game,  

1:25:30

1 hour, 25 minutes, 30 seconds

I’m looking forward to playing  it. I look a little bit like him.  

1:25:35

1 hour, 25 minutes, 35 seconds

Ladies and gentlemen, NVIDIA  RTX Spark laptops. Now,  

1:25:46

1 hour, 25 minutes, 46 seconds

thank you. I have too many things in my pocket. Okay, all right. This is the most amazing chip the  

1:26:00

1 hour, 26 minutes

world has ever built. This is the chip we built  in partnership with MediaTek. I think I saw Rick  

1:26:09

1 hour, 26 minutes, 9 seconds

earlier. This is a beautiful chip. This is a chip  that, frankly, would take 33 years to build. And  

1:26:19

1 hour, 26 minutes, 19 seconds

the reason for that is because 100% of NVIDIA  software runs here. If you want to run digital  

1:26:28

1 hour, 26 minutes, 28 seconds

biology, no problem. If you want to do seismic  processing, no problem. You want astrophysics,  

1:26:33

1 hour, 26 minutes, 33 seconds

no problem. Everything associated with CUDA — all  the physics, all the biology, all the genomics,  

1:26:38

1 hour, 26 minutes, 38 seconds

all the AI, all the computer graphics — no  problem. Every single application NVIDIA  

1:26:45

1 hour, 26 minutes, 45 seconds

has ever created, and every single application  that Windows has ever run. Microsoft and NVIDIA  

1:26:53

1 hour, 26 minutes, 53 seconds

meticulously optimized everything so that this  computer literally runs everything the world has  

1:27:01

1 hour, 27 minutes, 1 second

ever created. Plus, it now runs agents. It is  an incredible computer. I am so proud of it.  

1:27:15

1 hour, 27 minutes, 15 seconds

Okay. Now, I want you to keep that in mind  for the next video I’m going to show you.  

1:27:22

1 hour, 27 minutes, 22 seconds

Just imagine — everything here is going to run  on your PC. Now, that computer could have a  

1:27:28

1 hour, 27 minutes, 28 seconds

local Nemotron 3 Ultra model, or a Nemotron 3  model, or it could have Claude Code, or Codex,  

1:27:38

1 hour, 27 minutes, 38 seconds

or some other model in the cloud, or something  on the network. And it’s going to work and do   something amazing. Let’s play it. ________________  

1:27:48

1 hour, 27 minutes, 48 seconds

[VIDEO NARRATION] Every house starts as   an idea. Getting from idea to design  takes a myriad of tools, expertise,  

1:27:56

1 hour, 27 minutes, 56 seconds

and a lot of time. Now, an agent running  locally on RTX Spark can help me design  

1:28:04

1 hour, 28 minutes, 4 seconds

a house using the tools on my laptop — with an  OpenShell sandbox, running the Hermes harness,  

1:28:10

1 hour, 28 minutes, 10 seconds

connected to Claude Sonnet in the cloud. I select the site, share my concept sketches  

1:28:16

1 hour, 28 minutes, 16 seconds

and mood board of styles to inspire my design.  And the prompt — a text description of the  

1:28:22

1 hour, 28 minutes, 22 seconds

requirements and the design intent. My agent  goes to work. Using the tools on my laptop,  

1:28:31

1 hour, 28 minutes, 31 seconds

it opens Rhino and starts modeling the site,  shaping terrain, setbacks, and the building  

1:28:37

1 hour, 28 minutes, 37 seconds

envelope. Then it proposes building forms,  optimized for cost, comfort, and quality.  

1:28:45

1 hour, 28 minutes, 45 seconds

With the form defined, my agent generates  the interior layout. Walls, circulation,  

1:28:51

1 hour, 28 minutes, 51 seconds

rooms begin to take shape. I jump in whenever  I want to adjust, to change. Doors, windows,  

1:29:02

1 hour, 29 minutes, 2 seconds

and structural elements are placed automatically.  My agent detects its own mistakes and fixes them.  

1:29:11

1 hour, 29 minutes, 11 seconds

When I approve, the agent exports the model  from Rhino into Blender — materials and object  

1:29:17

1 hour, 29 minutes, 17 seconds

properties transfer with the design context  intact. I fine-tune the materials, get the look  

1:29:24

1 hour, 29 minutes, 24 seconds

just right. Then I pick the shots. Blender renders  the house. My agent, using generative AI with the  

1:29:31

1 hour, 29 minutes, 31 seconds

Flux 2 model, makes them photorealistic. Multiple  viewpoints, lighting conditions. What was once a  

1:29:38

1 hour, 29 minutes, 38 seconds

complex workflow is now guided and simplified  by my agent working on RTX Spark — designed at  

1:29:47

1 hour, 29 minutes, 47 seconds

the speed of imagination. ________________  

1:29:56

1 hour, 29 minutes, 56 seconds

[JENSEN] The developers are so excited   about this — it’s an incredible computer. All of  the acceleration, all the software capabilities  

1:30:05

1 hour, 30 minutes, 5 seconds

associated with it, working with every  developer to make it incredible for all of you.  

1:30:11

1 hour, 30 minutes, 11 seconds

The next one — Adobe. Incredible tool suite, used  by tens of millions of people around the world.  

1:30:18

1 hour, 30 minutes, 18 seconds

They have re-engineered the architecture,  the core of Adobe Photoshop and Premiere,  

1:30:24

1 hour, 30 minutes, 24 seconds

and they’ll release it for RTX Spark. It is twice  as fast. It’s already fast — now it’s going to be  

1:30:30

1 hour, 30 minutes, 30 seconds

twice as fast. And it’s also designed to  be agent-friendly — with its MCP server,  

1:30:40

1 hour, 30 minutes, 40 seconds

it can now interact with agents on your laptop. The number of partners that are so excited to  

bring RTX Spark to market is just incredible.  This is the first great PC reinvention in 40  

1:30:55

1 hour, 30 minutes, 55 seconds

years. And I’m just so happy that all of you  and the ecosystem around the world has joined  

1:31:02

1 hour, 31 minutes, 2 seconds

  1. This is basically everybody. Everybody  will support RTX Spark, and we’ll be building  

1:31:08

1 hour, 31 minutes, 8 seconds

incredibly smart and powerful and beautiful  laptops with all of you. Thank you very much.  

1:31:20

1 hour, 31 minutes, 20 seconds

But that’s not all. RTX Spark is a reinvention of  the laptop. But in fact, Microsoft and NVIDIA are  

1:31:30

1 hour, 31 minutes, 30 seconds

reinventing all of PC. And today we’re announcing  a whole new line — three revolutionary Windows  

1:31:39

1 hour, 31 minutes, 39 seconds

machines, covering desktop, laptop, and  workstation — all 100% Windows compatible,  

1:31:47

1 hour, 31 minutes, 47 seconds

100% CUDA, 100% NVIDIA AI Tensor Core.  Everything that runs on NVIDIA on all  

1:31:55

1 hour, 31 minutes, 55 seconds

these different platforms around the world, runs  here. This is the first completely re-engineered,  

1:32:04

1 hour, 32 minutes, 4 seconds

reinvented line of PCs in 40 years. Now, what’s really amazing is this. So  

1:32:10

1 hour, 32 minutes, 10 seconds

this is the RTX Spark laptop. This is the desktop.  This one’s from MSI — Joseph, this one’s yours.  

1:32:22

1 hour, 32 minutes, 22 seconds

Look how beautiful it is. This agent could run  24/7, meter-free. You can download your agent,  

1:32:31

1 hour, 32 minutes, 31 seconds

you can run it right here. It’s running all the  time. No meter anxiety. And it’s connected to  

1:32:44

1 hour, 32 minutes, 44 seconds

your whole house — connected to your laptop,  connected to your display, all the cameras,  

1:32:51

1 hour, 32 minutes, 51 seconds

your dryer, your water cooler, your water heater,  your everything, whatever you want — your security  

1:32:57

1 hour, 32 minutes, 57 seconds

system. All connected to this. And this becomes  your personal AI, your personal AI agent.  

1:33:04

1 hour, 33 minutes, 4 seconds

And it gets smarter and smarter and smarter  over time, because today we have Nemotron 3   Ultra. Tomorrow we have Nemotron 4. Then you  switch to Nemotron 5, then Nemotron 6, and we  

1:33:14

1 hour, 33 minutes, 14 seconds

just keep getting smarter, smarter, smarter. And  meanwhile, this is sitting at home, helping you do  

1:33:19

1 hour, 33 minutes, 19 seconds

things. If you want to book travel, no problem. And if you want an incredible system — this is DGX  

1:33:31

1 hour, 33 minutes, 31 seconds

Station. Compatible with Windows, runs everything  in Windows, and it has 768 gigabytes of memory.  

1:33:44

1 hour, 33 minutes, 44 seconds

And so you could run a trillion-parameter  model. This is unbelievable. 20 petaflops,  

1:33:51

1 hour, 33 minutes, 51 seconds

eight terabytes per second of memory  bandwidth, and this sits by your desk.  

1:33:58

1 hour, 33 minutes, 58 seconds

If you’re a developer of large language models,  a developer of agents — having this sit by your  

1:34:05

1 hour, 34 minutes, 5 seconds

desk gives you all the compute you need, and then  when you deploy it, you put it in the cloud.  

1:34:10

1 hour, 34 minutes, 10 seconds

Now, there’s something that, if you look at this  and think about this, something is happening here.  

1:34:18

1 hour, 34 minutes, 18 seconds

Remember 15, 20 years ago — we used to have  an idea called a phone. Today, when you think  

1:34:31

1 hour, 34 minutes, 31 seconds

about your phone, the one thing you don’t do  with it is make phone calls. You do just about  

1:34:40

1 hour, 34 minutes, 40 seconds

everything else. And so that phone means something  very different to you than a phone of the past.  

1:34:49

1 hour, 34 minutes, 49 seconds

I am certain what’s going to happen here is  that the PC, 10 years from now — the PC that  

1:34:55

1 hour, 34 minutes, 55 seconds

you think about today, a tool where you launch  applications, click and type — is going to be  

1:35:05

1 hour, 35 minutes, 5 seconds

completely different. Here’s my theory. I can  totally imagine, just as every house today  

1:35:12

1 hour, 35 minutes, 12 seconds

has a home theater, big TVs, lawn mowers,  dishwashers — I could totally imagine that  

1:35:24

1 hour, 35 minutes, 24 seconds

someday there’s actually an AI supercomputer  in your house. And it’s running all of your  

1:35:30

1 hour, 35 minutes, 30 seconds

agents. It’s running all of your assistants,  and they’re doing all kinds of things for you  

1:35:35

1 hour, 35 minutes, 35 seconds

all the time. You want AI agent computers  running in your house. And these, in time,  

1:35:51

1 hour, 35 minutes, 51 seconds

become a lot more like R2-D2 to you. It becomes  more like C-3PO to you than it feels like a PC.  

1:36:03

1 hour, 36 minutes, 3 seconds

There is no question — this reinvention of the  computer is as big a deal as the reinvention of  

1:36:09

1 hour, 36 minutes, 9 seconds

the phone into what we now know as the smartphone.  And so, this is the beginning of that journey.  

1:36:15

1 hour, 36 minutes, 15 seconds

This is the beginning of a new line. We have a  roadmap for this — a brand new product family.  

1:36:23

1 hour, 36 minutes, 23 seconds

Every single generation of architecture, we will  have a desktop, a laptop, and a workstation.  

1:36:33

1 hour, 36 minutes, 33 seconds

And the thing that I’m just incredibly pleased  and incredibly honored by is that 100% of the  

1:36:40

1 hour, 36 minutes, 40 seconds

world’s PC industry has joined us to reinvent  the PC. A new line, a new beginning. Thank you.  

1:37:01

1 hour, 37 minutes, 1 second

As you know, agentic AI is just a digital  robot. It understands, it reasons, it plans,  

1:37:13

1 hour, 37 minutes, 13 seconds

and it acts and uses tools. Agentic AI is going to  run across all of these computers. We’re working  

1:37:24

1 hour, 37 minutes, 24 seconds

on humanoid robotics computers, robotics computers  of all kinds. We’re working on self-driving car  

1:37:30

1 hour, 37 minutes, 30 seconds

computers. We’re working on satellites. You  have GeForce, which has Tensor Cores

COSMOS 3 Model for AGENTIC ROBOTIC AI & PHYICS AI WORLDS (Also Axios review)

on humanoid robotics computers, robotics computers  of all kinds. We’re working on self-driving car  

1:37:30

1 hour, 37 minutes, 30 seconds

computers. We’re working on satellites. You  have GeForce, which has Tensor Cores. I just  

1:37:36

1 hour, 37 minutes, 36 seconds

talked about a whole new line of PCs. Agriculture  equipment, manufacturing equipment, heavy industry  

1:37:42

1 hour, 37 minutes, 42 seconds

equipment — all will be agentic. Even your base  stations — the radio stations of the future — are  

1:37:54

1 hour, 37 minutes, 54 seconds

going to be agentic, understanding traffic and  thinking about how to coordinate with other base  

1:38:01

1 hour, 38 minutes, 1 second

stations so that you can use as little energy as  possible, increasing the spectral efficiency.  

1:38:10

1 hour, 38 minutes, 10 seconds

And so everything will run agents. Today, NVIDIA  is largely in the center. But I am pretty certain  

1:38:18

1 hour, 38 minutes, 18 seconds

that there will be tens of billions, hundreds of  billions over time, of agentic systems and agentic  

1:38:26

1 hour, 38 minutes, 26 seconds

computers running around the world. The biggest problem is data. In the  

1:38:32

1 hour, 38 minutes, 32 seconds

case of language models, all the English and  all the language that we have on the internet   that we train on was from the perspective  of us — we wrote it and we’re reading it.  

1:38:42

1 hour, 38 minutes, 42 seconds

However, in order to create data for AI robotics,  it has to be from the perspective of the robot.  

1:38:52

1 hour, 38 minutes, 52 seconds

And most of the world’s video data is from  a third person, not first person. And so,  

1:38:58

1 hour, 38 minutes, 58 seconds

agentic systems, robotic systems, physical  AI — the data is the hardest problem.  

1:39:06

1 hour, 39 minutes, 6 seconds

You’ve seen us move up this ladder. We started  with teleoperation, which is basically human  

1:39:11

1 hour, 39 minutes, 11 seconds

demonstration — no different than the big  breakthrough of reinforcement learning from human  

1:39:17

1 hour, 39 minutes, 17 seconds

feedback. Then we use simulation — this is  where Omniverse comes in, no different than  

1:39:23

1 hour, 39 minutes, 23 seconds

reinforcement learning with verifiable rewards.  We use these systems to bootstrap the AI model,  

1:39:34

1 hour, 39 minutes, 34 seconds

the physical AI model. Eventually we’re able  to learn from third-person data, reprojecting  

1:39:41

1 hour, 39 minutes, 41 seconds

into first person. And now, eventually, through  bootstrapping, we have a World Foundation Model  

1:39:48

1 hour, 39 minutes, 48 seconds

that can understand the physical world from any  perspective you want — third person, first person,  

1:39:56

1 hour, 39 minutes, 56 seconds

inside and out. This is a big breakthrough. And today, we’re announcing Cosmos 3.  

1:40:07

1 hour, 40 minutes, 7 seconds

Cosmos 3 is the frontier of physical AI. We are at  the frontier with language models — there are so  

1:40:16

1 hour, 40 minutes, 16 seconds

many people working on it. However, in physical  AI, we are absolutely the world’s best. I am so  

1:40:23

1 hour, 40 minutes, 23 seconds

proud of the team for doing this. This is the  Foundation Model for all of your work. Whenever  

1:40:28

1 hour, 40 minutes, 28 seconds

you want to create a robot — a factory robot, a  robot that works in a factory, any kind of robot  

1:40:35

1 hour, 40 minutes, 35 seconds

that involves the physical world — you now have  a companion in Cosmos 3 that can understand and  

1:40:44

1 hour, 40 minutes, 44 seconds

reason. It can generate, it can simulate, and in  the loop it can even be the policy itself. It is  

1:40:52

1 hour, 40 minutes, 52 seconds

at the top of leaderboards all over the world.  I am incredibly proud of Cosmos, and today we’re   announcing Cosmos 3. Let’s take a look. ________________  

1:41:01

1 hour, 41 minutes, 1 second

[VIDEO NARRATION] The real world is infinite   and unpredictable. Physical AI needs data,  but real-world data is impossible to scale.  

1:41:10

1 hour, 41 minutes, 10 seconds

For physical AI, compute is data. This is Cosmos  — an open frontier omni-model for physical AI,  

1:41:20

1 hour, 41 minutes, 20 seconds

built on a new Mixture of Transformers  architecture. Pixels, action, sound,   and language flow into the  autoregressive transformer,  

1:41:28

1 hour, 41 minutes, 28 seconds

which reasons, plans, and instructs the diffusion  transformer, which generates what comes next.  

1:41:35

1 hour, 41 minutes, 35 seconds

Developers post-train Cosmos across  embodiments and use cases. As a VLM,  

1:41:41

1 hour, 41 minutes, 41 seconds

Cosmos watches the physical world, understands  what’s happening, describing scenes and flagging  

1:41:48

1 hour, 41 minutes, 48 seconds

what matters. As a world model, Cosmos generates  physics-accurate synthetic video from an image,  

1:41:56

1 hour, 41 minutes, 56 seconds

text, or video. As a simulator, Cosmos closes  the loop for policy training and evaluation.  

1:42:03

1 hour, 42 minutes, 3 seconds

And as the foundation of NVIDIA OmniDreams  — an action-conditioned world model — Cosmos  

1:42:09

1 hour, 42 minutes, 9 seconds

predicts the future frame by frame. Post-train Cosmos and it becomes a world  

1:42:16

1 hour, 42 minutes, 16 seconds

action model — perceiving, reasoning, planning,  generating actions — for robots of every kind,  

1:42:26

1 hour, 42 minutes, 26 seconds

for everything that moves. A new  kind of data. A new kind of teacher,  

1:42:33

1 hour, 42 minutes, 33 seconds

generated by compute. Cosmos — the foundation  for developers in the age of physical AI.  

1:42:55

1 hour, 42 minutes, 55 seconds

________________ [JENSEN]   Text data plus compute gives you AI. Now that we  have AI, compute is data. And so, use Cosmos 3,  

1:43:07

1 hour, 43 minutes, 7 seconds

train a whole bunch of models. Cosmos is such  an incredible open model system — it’s exactly   the same as Nemotron. We open the model, we open  the data, and we even open how we trained it so  

1:43:17

1 hour, 43 minutes, 17 seconds

that you can enhance it for yourself and  turn Cosmos into your proprietary model.   We have such incredible partners working  with us in so many different industries.  

1:43:27

1 hour, 43 minutes, 27 seconds

Now, the model itself is, of course, the most  understandable part of the AI stack. But the AI  

1:43:34

1 hour, 43 minutes, 34 seconds

stack is very complicated — it has generators, the  model, simulators, and the runtime. Just as it is  

1:43:44

1 hour, 43 minutes, 44 seconds

for agentic systems, cars are essentially physical  AI — agentic robots that are autonomous vehicles  

1:43:54

1 hour, 43 minutes, 54 seconds

— and they have this complicated stack too. Today we’re announcing Alpamayo 2 Super, an  

1:44:00

1 hour, 44 minutes

open model for self-driving cars. We’re  working with car companies across the world.  

1:44:07

1 hour, 44 minutes, 7 seconds

If you look at the brands that have signed up  for NVIDIA DRIVE Hyperion — that are building  

1:44:11

1 hour, 44 minutes, 11 seconds

NVIDIA DRIVE Hyperion cars — this represents  about 80% of the world’s car manufacturers.  

1:44:23

1 hour, 44 minutes, 23 seconds

We are going to have a whole lot of NVIDIA  DRIVE Hyperion systems that are able to run  

1:44:28

1 hour, 44 minutes, 28 seconds

Alpamayo 2 Super, or anybody else’s NVIDIA  stack. We are also connected into mobility  

1:44:34

1 hour, 44 minutes, 34 seconds

services — approximately 97% of the world’s  mobility services are connecting with us — so  

1:44:40

1 hour, 44 minutes, 40 seconds

that when we deploy Alpamayo 2 Super on the DRIVE  Hyperion runtime, with the Halos operating system,  

1:44:48

1 hour, 44 minutes, 48 seconds

we will be able to connect to all of these  services across the world. Let’s take a look.   ________________ [VIDEO NARRATION]  

1:44:55

1 hour, 44 minutes, 55 seconds

Hey Mercedes, let’s go to  my favorite sandwich shop.   Routing to your destination. Lane is clear,  pulling out to start drive. Nudge left due  

1:45:06

1 hour, 45 minutes, 6 seconds

to the stationary lead vehicle ahead blocking  our lane. Slow down to stop at the stop sign,  

1:45:12

1 hour, 45 minutes, 12 seconds

controlling the intersection. Stop to yield to  the pedestrians. Since the person is in our lane,  

1:45:18

1 hour, 45 minutes, 18 seconds

yield for the cutting vehicle from the left. Nudge  left to clear the stopped vehicle blocking on the   right. Keep distance to the cutting vehicle  since it is merging into our lane. There is a  

1:45:28

1 hour, 45 minutes, 28 seconds

lane stream crossing ahead. Stop to keep distance,  decelerate. Stop at the stop sign, intersection is  

1:45:37

1 hour, 45 minutes, 37 seconds

controlled. Stop, yield to the cross traffic. Keep  distance to the truck. Nudge left due to the truck  

1:45:41

1 hour, 45 minutes, 41 seconds

blocking on the left side of our lane. Your destination is on the right.  

1:45:57

1 hour, 45 minutes, 57 seconds

Alpamayo 2 Super — the world’s first reasoning   autonomous vehicle. ________________  

1:46:05

1 hour, 46 minutes, 5 seconds

[JENSEN] If you let it talk all the time, it will drive   you crazy. But we’re very happy that it’s talking  to itself all the time. That’s called thinking.  

1:46:17

1 hour, 46 minutes, 17 seconds

And so Alpamayo 2 Super is a reasoning car. The technology that we’ve created also applies   to humanoids. Of course, there are many  new breakthroughs that have to happen.  

1:46:26

1 hour, 46 minutes, 26 seconds

NVIDIA Isaac GR00T is our humanoid robotics  stack — model, data generation, simulation,  

1:46:38

1 hour, 46 minutes, 38 seconds

the runtime including the operating system.  This represents the GR00T platform, the Isaac  

1:46:47

1 hour, 46 minutes, 47 seconds

GR00T platform. Every one of our systems — the  exact same architecture — whether it’s an agentic  

1:46:53

1 hour, 46 minutes, 53 seconds

system for the cloud, an agentic system for the  PC, a robotic system for a self-driving car,  

1:47:00

1 hour, 47 minutes

a robotic system for a humanoid robot — all  the same. And of course, in every single case,  

1:47:06

1 hour, 47 minutes, 6 seconds

we build everything completely. We built  everything vertically, completely integrated  

1:47:15

1 hour, 47 minutes, 15 seconds

with co-design, extreme co-design. And  then when we open it up for everybody   to use — whichever part you like, whatever  you want to use, we even help you modify.  

1:47:26

1 hour, 47 minutes, 26 seconds

But the one thing that is missing is we need  a reference platform for robotic systems.  

1:47:33

1 hour, 47 minutes, 33 seconds

These robotic systems are so complicated  — so many motors, so many sensors,   so fragile — and yet we need to have a way to  deliver these reference platforms, just like we  

1:47:44

1 hour, 47 minutes, 44 seconds

do with PCs and DGXs and clouds and self-driving  cars. We now are going to do it for robots. Today,  

1:47:51

1 hour, 47 minutes, 51 seconds

we’re announcing the NVIDIA Isaac GR00T reference  humanoid robot — fully integrated. 25 degrees of  

1:47:59

1 hour, 47 minutes, 59 seconds

freedom on each hand, made by Sharpa. 31 degrees  of freedom on the robot, 6 feet, 150 pounds. Just  

1:48:09

1 hour, 48 minutes, 9 seconds

like me. The first number is shorter, the second  number is bigger. Otherwise, pretty close.  

1:48:20

1 hour, 48 minutes, 20 seconds

And this platform runs the new Jetson Thor and  our entire software stack — data generation stack,  

data simulation stack, the runtime —  all integrated into a robot designed  

for everyone to use. We built this for  higher education and university researchers,  

because for them to build this themselves is  insanely hard. And so let’s take a look at that.   ________________ [VIDEO NARRATION]  

1:48:46

1 hour, 48 minutes, 46 seconds

The next leap in AI is general-purpose  robots — humanoids. But building one is  

1:48:51

1 hour, 48 minutes, 51 seconds

hard. Every team starts from scratch, stitching  together simulators, teleoperation systems,  

1:48:57

1 hour, 48 minutes, 57 seconds

data pipelines, and training infrastructure.  Months of setup before research can start.  

1:49:03

1 hour, 49 minutes, 3 seconds

NVIDIA Isaac GR00T — an open development  platform for humanoid robots. Open models,  

simulation and training libraries, and data  generators. Plus, the robot computer — fully  

pre-configured, ready to go in hours. First, set up the simulation environment  

in Isaac Lab. Capture demonstrations with  Isaac Teleoperation on a real or simulated  

robot. Generate synthetic data with Omniverse and  Cosmos, scaling one demonstration into thousands.  Train policies, evaluate them in Isaac Lab Arena.  Deploy through Isaac ROS, running on Jetson Thor.  

Every element is modular and open — use ours  or swap in your own. GR00T is powering robotics  

research across every discipline for every  domain, from research labs to factory floors.  

One open platform. And now, a new edition: Isaac  GR00T reference design robots — built on NVIDIA’s   open platform, ready for frontier research, for  any lab, anywhere. The age of robotics starts  here. NVIDIA Isaac GR00T. ________________  

1:50:48

1 hour, 50 minutes, 48 seconds

[JENSEN] So many robots. Well, we can just  

1:50:56

1 hour, 50 minutes, 56 seconds

let everybody who’s working on robots in the world  — let me tell you what I told you.The computer  industry has been completely changed. In the  last six months, everything changed. Everything  changed because agents were realized, and it  converged with the latest frontier models, and  it made possible for AI to now do useful work. The computing pattern will repeat over and over  

and over again. This computing pattern — an agent  that has a model, a harness that uses tools with  

skills, and runs in a runtime — that runtime  depends on whether it’s in a cloud or on-prem,  

on a PC or a robot. But the computing pattern  is exactly the same for all of them. You will  

use different harnesses because of your  preference. You’ll use different models   because of your preference. You will improve  them for your proprietary use. You would  

create super agents that you can offer to  other people to help them do their work.  

This agentic platform, this agentic pattern —  NVIDIA has an Enterprise AI Toolkit. This is a  

1:52:06

1 hour, 52 minutes, 6 seconds

wonderful way for all of you to engage with AI,  and for us it’s a wonderful growth opportunity.  

Vera Rubin is in full production. Whereas Grace  Blackwell was created to process AI — particularly  

inference — Vera Rubin was created to run agents.  It is in full production. It is much, much more  

than a GPU. It is an entire, disaggregated,  distributed agent processing system.  

1:52:35

NVIDIA has really become an infrastructure company  — not just a GPU company, not just a systems  

1:52:41

1 hour, 52 minutes, 41 seconds

company, but an infrastructure company to help you  generate the maximum revenue, the maximum profit,  

1:52:48

1 hour, 52 minutes, 48 seconds

and to get there as soon as possible. The agentic world — this new way of  

1:52:54

1 hour, 52 minutes, 54 seconds

computing — where you build CPUs for agents,  not for people. CPUs for agents have their  

1:53:01

1 hour, 53 minutes, 1 second

own special requirements, and NVIDIA Vera is  revolutionary. I’m so happy about its ramp.  

1:53:08

1 hour, 53 minutes, 8 seconds

The orders are already in. It’s going  to be the fastest and most successful   product launch in our company’s history. NVIDIA and Microsoft have created a whole  

1:53:18

1 hour, 53 minutes, 18 seconds

new line of PCs. This is a new beginning, and of  course, the exact same agentic computing pattern  

1:53:26

1 hour, 53 minutes, 26 seconds

that I just described is also going to run on  all kinds of devices. I mentioned PCs, but in  

1:53:34

1 hour, 53 minutes, 34 seconds

the future it’ll be robots and satellites and base  stations and factories — in the cloud, on-prem,  

1:53:41

1 hour, 53 minutes, 41 seconds

at the edge. This agentic computing pattern will  be replicated in computers all over the world.  

1:53:50

1 hour, 53 minutes, 50 seconds

How we think about the personal computer will very  likely change. I want to thank all of you for your  

1:53:56

1 hour, 53 minutes, 56 seconds

partnership, your friendship. We couldn’t be here  without everything that we do together. I am so  

1:54:02

1 hour, 54 minutes, 2 seconds

proud of how successful you’ve been this last  year. The next year is going to be even more. I  

1:54:09

1 hour, 54 minutes, 9 seconds

have one more thing for you. Let’s take a look. ________________  

1:54:32

1 hour, 54 minutes, 32 seconds

[VIDEO NARRATION] [Closing performance — partial lyrics as  

1:54:38

1 hour, 54 minutes, 38 seconds

captured:] Useful AI has arrived,  

1:54:45

1 hour, 54 minutes, 45 seconds

agents working by your side. Can’t miss the  things we said today — we’re gonna break it  

1:54:51

1 hour, 54 minutes, 51 seconds

all down for you, Taipei. Agents — widely  misunderstood, only movie stars had them  

1:54:56

1 hour, 54 minutes, 56 seconds

in Hollywood. Now they’re your coworkers, making  dreams come true, building companies from living  

1:55:02

1 hour, 55 minutes, 2 seconds

rooms. There’s so much confusion — we hear you.  That’s why we created Vera, gonna show it’s true.  

1:55:09

1 hour, 55 minutes, 9 seconds

The cheapest token coming through. Ten times  throughput — more for special agents, double  

1:55:17

1 hour, 55 minutes, 17 seconds

down. Vera keeps agent memory true. Now let’s talk  about it — CPU. Fifty percent faster. Not for x86.  

1:55:25

1 hour, 55 minutes, 25 seconds

It’s built for agents. Everyone’s welcome to the  NVIDIA party. Vera Rubin in full production. 5X,  

1:55:40

1 hour, 55 minutes, 40 seconds

faster work is done. OpenShell, keep the sandbox  tight. Here’s a five-layer cake, make no mistake.  

1:55:58

1 hour, 55 minutes, 58 seconds

MaxLPS keeps power lean. Optimized for you —  so you can have your cake and eat it too. PC  

1:56:11

1 hour, 56 minutes, 11 seconds

reinvention — 40 years. Running anywhere Windows  goes. Turning compute into synthetic data,  

1:56:29

1 hour, 56 minutes, 29 seconds

reasoning through, understanding the world  like people do. It’s how they learn to move.  

1:57:12

1 hour, 57 minutes, 12 seconds

The future’s here — come see what’s next. ________________  

1:57:19

1 hour, 57 minutes, 19 seconds

[JENSEN] Thank you, Taiwan. Welcome to Computex.  

1:57:36

1 hour, 57 minutes, 36 seconds

Have a great time, everyone.  Thanks for an amazing year.   Thank you for all your friendship  and support. Thank you. Take care.

June 2026 Axios review of cosmos

Nvidia unveiled Cosmos 3, an open AI world model designed to help robots, autonomous vehicles and other physical systems better understand and predict real-world environments.

Why it matters: Nvidia is continuing its move beyond chips into AI models and software, positioning itself to become a foundational platform for physical AI development.

Driving the news: Nvidia says it trained Cosmos 3 on 20 trillion tokens of multimodal data, including nearly a billion images, 400 million real and synthetic videos, ambient audio, text and action data from humans and robots.

  • That action data is what makes Cosmos different from a regular video generator. It's meant to model how machines move, not just how scenes look, Ming-Yu Liu, VP of Nvidia's Cosmos Lab, told Axios. Autonomous actions are key.
  • Developers can use Cosmos 3 to simulate actions in physical environments, then build task-specific models for robots and other machines on top of it.
  • Cosmos 3 is designed to generate action data — such as robot joint angles, gripper positions and trajectories — that can help train machines to navigate and manipulate the physical world.

Between the lines: Cosmos is an open model, similar to its early Nemotron family, making it easier for hardware makers to customize Cosmos to their needs and ensure that future versions more closely align to the needs of the industry, Liu said.

  • Nvidia is also establishing a coalition of companies supporting the effort. Initial partners include Agile Robots, Black Forest Labs and Runway.
  • Nvidia says Cosmos can create rare or dangerous scenarios — such as robot collisions or unusual road events — that are difficult, expensive or unsafe to capture repeatedly.

Zoom in: Nvidia is releasing two versions immediately: a "super" model for tasks requiring high physics accuracy, such as training robots and autonomous vehicles, and a "nano" model that can produce results in fractions of a second.

Gemini adds this on supplementary platform Omiidreams

NVIDIA OmniDreams is a generative world model designed to create photorealistic, closed-loop simulation environments for autonomous vehicles (AVs) and robotaxis. [1]
Introduced alongside Cosmos 3 and the Alpamayo 2 Super reasoning models at Computex, "OmniDreams" functions as a physical AI agent skill that essentially lets a computer "dream up" highly specific real-world driving environments. [1, 2, 3, 4]
The model serves several distinct purposes within NVIDIA's physical AI ecosystem: [1]
1. Simulating "Long-Tail" Edge Cases
The hardest part of training self-driving cars is preparing them for rare, dangerous, or unusual road conditions (known as "long-tail" scenarios) that rarely occur in real life. OmniDreams generates these hyper-specific edge cases at scale—such as extreme weather, blinding glare, or unexpected road obstacles—allowing AV models to experience them safely in simulation. [1, 2]
2. Closed-Loop Training
Unlike static video generation, OmniDreams operates in a closed loop. If the autonomous vehicle decides to steer left or slam on the brakes, OmniDreams dynamically alters the simulated environment in real time to reflect the consequences of that driving decision. [1, 2, 3]
3. High-Fidelity Synthetic Data Generation
Trained on top of the physics and spatial reasoning frameworks of the Cosmos platform, OmniDreams takes a starting image or sequence of actions and renders a photorealistic, physically plausible world. This creates flawless synthetic sensor data that developers use to train and validate self-driving AI before the car ever touches physical pavement. [1, 2, 3, 4]
Would you like to know more about the AlpaGym reinforcement learning framework that pairs with OmniDreams, or how developers use Omniverse NuRec to turn real-world fleet footage into these 3D simulation spaces? [1, 2]

Some of nvidia's expanding log of robot partnerships - gemini

NVIDIA has built an massive global robotics ecosystem spanning humanoid creators, industrial giants, open-source AI communities, and logistics pioneers. Jensen Huang’s vision of "Physical AI" relies on a "three-computer architecture" combining the NVIDIA Omniverse (for simulation), NVIDIA DGX (for AI training), and NVIDIA Jetson Thor / Blackwell chips (for on-device reasoning). [1, 2, 3, 4, 5]

Major global robotics partnerships leveraging the NVIDIA Isaac Platform are categorized below: [1]

Humanoid Robotics & Physical AI Startups

NVIDIA's Isaac GR00T foundation model and Jetson Thor processors have been rapidly adopted by the world's leading humanoid developers: [1, 2, 3, 4]

  • Unitree Robotics (China): Announced as a primary partner for NVIDIA’s first general humanoid research platform. The 6-foot-tall Unitree H2 humanoid robot uses Jetson Thor and Blackwell chips to run Isaac GR00T. [1]
  • Sharpa (Singapore): Collaborating directly alongside the Unitree initiative to supply high-dexterity mechanical hands for AI experimentation. [1, 2]
  • Boston Dynamics (USA): Integrating Jetson Thor into its advanced humanoids to drastically enhance navigation and complex manipulation. [1]
  • Figure AI (USA): Developing collaborative humanoids for the commercial workforce trained inside Isaac simulation environments. [1, 2]
  • Agility Robotics (USA): Utilizing Isaac Sim and Isaac Lab to train and simulate their Digit humanoid workforce. [1, 2]
  • NEURA Robotics (Germany): Launching Porsche-designed, dexterous Gen 3 humanoid systems built directly on NVIDIA's edge computing architecture. [1, 2]
  • Other Notable Humanoid Partners: 1X, AGIBOT, Agile Robots, Fourier, Mentee Robotics, Noble Machines, and XPENG Robotics. [1, 2]

Industrial Automation Giants

The world's largest industrial robotics companies—representing millions of active machinery units globally—rely on NVIDIA Omniverse libraries for creating high-fidelity factory digital twins and virtual testing: [1, 2, 3]

  • FANUC (Japan): Integrating Isaac simulation into their virtual factory lines and utilizing Jetson edge-AI modules within their robot controllers.
  • ABB Robotics (Switzerland): Scaling physical AI deployments across industrial automation via NVIDIA Cosmos and Isaac tools.
  • YASKAWA (Japan): Co-developing advanced AI workloads for heavy industrial manufacturing.
  • KUKA (Germany): Implementing NVIDIA accelerated computing at the edge for industrial robotic controllers.
  • Universal Robots (Denmark): Powering cobots (collaborative robots) with the UR AI Accelerator, lowering barrier to deploy smart automation. [1, 2, 3, 4]

Logistics, Mobility & Last-Mile Delivery

NVIDIA's Isaac Perceptor framework serves as the primary visual engine for Autonomous Mobile Robots (AMRs) in large-scale supply chains: [1]

  • Serve Robotics (USA): Sidewalk delivery robots powered by strategic NVIDIA investment and AI, navigating last-mile food delivery.
  • Amazon Robotics (USA): Simulating massive fleet deployments and warehouse design through the industrial Omniverse blueprint.
  • Foxconn (Taiwan): Actively building fully simulated autonomous factory digital twins using Isaac blueprints to run fleets of mobile AI assembly robots.
  • KION Group & BYD: Deploying visual-AI guidance systems in industrial forklifts and reach trucks using the NVIDIA Nova Orin DevKit. [1, 2, 3, 4, 5]

Open Source AI & Software Foundations

To guarantee developer adoption, NVIDIA has interconnected its proprietary tech stack with the largest open AI communities:

  • Hugging Face: Integrating NVIDIA Isaac and GR00T platforms directly into the LeRobot open-source robotics framework, bridges 2 million robotics experts with 13 million AI model builders. [1]
  • Google DeepMind & Disney Research: Co-developed the Newton physics simulation engine, an open-source, GPU-accelerated engine fully compatible with NVIDIA's learning frameworks. [1]

Would you like to focus on a specific region (such as NVIDIA's growing partnerships in South Korea and Europe), or dive deeper into the technical specifications of the Jetson Thor chipsets powering these humanoids? [1, 2]

20 sites

  • NVIDIA and Global Robotics Leaders Take Physical AI to the Real ...

Mar 16, 2026 — NVIDIA and Global Robotics Leaders Take Physical AI to the Real World. ... News Summary: Physical AI leaders across robot brain de...

NVIDIA Newsroom

  • NVIDIA and Global Robotics Leaders Take Physical AI to the Real ...

Mar 16, 2026 — Helping Tomorrow's Physical AI Pioneers Get Off the Ground NVIDIA is committed to ensuring that the tools for physical AI are acce...

NVIDIA Investor Relations

  • NVIDIA Releases New Physical AI Models as Global Partners Unveil ...

Jan 5, 2026 — Humanoid Robot Developers Adopt NVIDIA Jetson Thor NVIDIA Jetson Thor meets the massive computing requirements for humanoid robots...

NVIDIA Investor Relations

Show all

You said: Is there longer list, jensen computex talk staged local robots building with modularisaac groot platform including Linda Hyundai Universal Rainbow Nobel holiday fauna unitree,

Yes, Jensen Huang's keynote at Computex 2026 featured a massive, unified lineup of local and global robotics hardware partners. During the presentation, he officially launched the NVIDIA Isaac GR00T Humanoid Reference Robot platform. This modular blueprint allows third-party developers to easily swap out components like robotic limbs, torsos, and chassis. [1, 2, 3, 4, 5, 6]

By analyzing the specific phonetic and auto-corrected names from your list, here is the official, expanded breakdown of the major global partners building on the modular Isaac GR00T platform: [1]

The Core Reference Humanoid Partners

  • Unitree Robotics (China): Serving as the anchor body for the official reference platform. Their nearly 6-foot-tall Unitree H2 (and H2 Plus) humanoid serves as the standardized physical frame sold directly to researchers. [1, 2, 3]
  • Sharpa (Singapore): Contributed the tactile, five-fingered dexterous hands (the Sharpa Wave) to the modular reference kit. [1, 2]
  • Linda / LINAGORA (France): Part of NVIDIA's expanded local language and sovereign AI initiatives. They utilize the open model family to build local agentic frameworks for physical automation. [1]

Global Humanoid & Industrial Titans (Your List Decoded)

  • Hyundai / Boston Dynamics (South Korea/USA): Hyundai-owned Boston Dynamics is actively integrating NVIDIA's Jetson Thor computing brains to power its next-generation fully electric Atlas humanoids. [1, 2]
  • Universal Robots (Denmark): Expanding on their physical AI footprint, they use modular NVIDIA Isaac frameworks to build automated arms that can be trained natively inside simulation. [1]
  • Rainbow Robotics (South Korea): A major humanoid developer that officially adopted NVIDIA's Jetson Thor platform to transition their commercial bipedal systems into modular physical AI agents. [1, 2, 3]
  • Noble Machines (USA): Part of the wave of modular humanoid startups engineering specialized industrial manipulation systems on top of the open Isaac GR00T software. [1]
  • Fauna / FANUC (Japan): FANUC, the industrial automation titan, utilizes NVIDIA Cosmos world models and Isaac simulation tools to orchestrate heavy factory robotics and mobile logistics bots. [1, 2]

Elite Research Institutions Deploying the Reference Design [1]

Instead of just working with commercial brands, NVIDIA shipped this unified platform to elite academic groups to build decentralized skills: [1, 2, 3]

  • Stanford Robotics Center
  • ETH Zurich (Switzerland)
  • UC San Diego (Advanced Robotics & Controls Lab)
  • Allen Institute for AI (Ai2) [1, 2]

V10+3 determining whether AI creates jobs

V10 Taiwan, USA ,UK, Japan  S Korea, India, UAE, Saudi, Germany, France +3 biggest but nor with Vera,

  • China - the greatest number of engineering brains not to have Vera access
  • Singapore small but arguably greatest sponsor of cultural respect eg sponsors all asean language llm,
  • Musk while not a nation - lead partner of Jensen's leap in AI from autonomous cars to space including starlink

V10 current expected premier league of nations investing in Vera Rubin AI factories

Critical survey questions to V10_2

- which will apply their AI sovereignty to layer 5 AI and job creation 

Intel supplied by Grok-X as at 1 June 2026

Taiwan is positioned as a critical player in NVIDIA’s Vera Rubin (often referred to as the Vera/Rubin platform: Vera CPU + Rubin GPU) rollout, thanks to its dominance in advanced semiconductor manufacturing via TSMC and a strong ecosystem of partners (Foxconn, Pegatron, Quanta, Wistron, etc.). Over 150 NVIDIA ecosystem partners in Taiwan alone are ramping production for Rubin infrastructure across hundreds of factories.
blogs.nvidia.com
Taiwan is actively embracing Layer 5 applications (AI tools, industry-specific solutions, smart manufacturing, healthcare AI, etc.) through government initiatives like the AI New Ten Major Construction plan, free GPU resources for startups, and focus on practical adoption in manufacturing, healthcare, and smart cities. Its strengths in hardware-software integration make it well-suited for joyful, productive Layer 5 scaling.
moda.gov.tw
Interpreting "V10" and Ranking Top Nations
“Vera nations” and “V10” likely refer to the top 10 countries most critical to or investing heavily in NVIDIA’s Vera Rubin-era AI infrastructure (chips, data centers, sovereign AI, and Layer 5 apps). No formal “V10” alliance exists, but countries form partnerships through NVIDIA collaborations, hyperscaler builds, sovereign AI programs, and supply chains.
Ranking is based on a mix of:
AI investment & private funding
Data center/compute scale
Semiconductor/infrastructure role
NVIDIA partnerships & Rubin readiness
Layer 5 adoption potential (applications, jobs, community solutions)
Approximate 2025/2026 Top 10 Ranking (Likelihood/Impact for Vera Rubin Success):
United States — Dominant in private investment (~$470B+), models (OpenAI, Anthropic, etc.), hyperscalers (Microsoft, Google, Amazon, Meta), and domestic manufacturing pushes. Major Rubin deployments via NVIDIA partnerships. Strong Layer 5 ecosystem (coding tools, enterprise apps). Highest overall impact.
sphericalinsights.com
Taiwan — Critical manufacturing chokepoint (~90% advanced chips via TSMC). Hundreds of partners building Rubin systems. Strong push into Layer 5 applications (manufacturing AI, healthcare). High likelihood of joyful adoption as you noted.
blogs.nvidia.com
China — Massive scale, government funds, and domestic alternatives, though export restrictions limit latest NVIDIA access (capacity shifting toward Rubin for non-China markets). Focus on self-reliant Layer 5 in manufacturing and apps.
rand.org
United Kingdom — Large NVIDIA-backed AI factories (tens of thousands of GPUs), sovereign AI push, research, and partnerships (e.g., with Nscale). Growing Layer 5 in public services and innovation.
nvidianews.nvidia.com
South Korea — Semiconductor/memory leader (Samsung, SK Hynix for HBM), robotics, and AI chips. Strong industrial applications.
franklintempletonglobal.com
India — Rapid infrastructure growth, massive talent pool, data centers, and AI for public good (healthcare, agriculture, education). NVIDIA partnerships expanding. High Layer 5 scaling potential.
sphericalinsights.com
UAE — Leading AI adoption rates, sovereign AI (G42, massive GPU clusters), and energy/resources for data centers. Fast mover on applications.
visualcapitalist.com
Saudi Arabia — Huge Project Transcendence ($100B+), data center builds, and sovereign AI ambitions.
hai.stanford.edu
Japan — Tech giants (SoftBank, etc.), manufacturing strengths, and growing AI infrastructure investments.
x.com
Germany/France (Europe leaders) or Singapore/Canada/Israel — Strong in targeted investments, industrial AI (Germany), ethical frameworks/research (France/Canada), or adoption/innovation hubs (Singapore/Israel). France and Germany feature in top government commitments.
franklintempletonglobal.com
How They Partner as a "V10"-Style Network
These nations don’t form a rigid bloc but interconnect through NVIDIA’s global ecosystem, hyperscalers (Microsoft, Google, etc.), and supply chains:
Supply Chain Core: Taiwan + South Korea + US (design) + Japan (materials/equipment) enable Rubin production.
Build & Deploy: US, UK, UAE, Saudi, India build large AI factories (“AI factories” for agentic AI).
Applications (Layer 5): All are investing in sector-specific uses (Taiwan/ SK manufacturing; India public services; US/UK/France innovation & enterprise tools like coding assistants).
Geopolitical Flavor: US allies (Taiwan, SK, Japan, UK, India, UAE) collaborate closely on secure supply chains. China pursues parallel sovereign path. Europe balances investment with regulation.
Cross-Border Wins: NVIDIA sovereign AI clouds, joint data centers, talent exchanges, and open MGX designs allow smaller players to participate. Microsoft/IBM-style enterprise partnerships extend Layer 5 tools globally.
Taiwan’s joyful Layer 5 embrace could shine in practical apps (e.g., AI for semiconductors, smart factories, education), inspiring others. Success depends on education, trust, and community projects as you emphasized—parents, schools, and builders experimenting with tools like Cursor/Copilot on top of the full stack.
The real test is turning infrastructure (lower layers) into widespread human benefit at Layer 5. Which country or partnership angle interests you most for deeper details?
Apart from Taiwan I wonder which country cares for Jensen and Taiwan must. But as well as 10 named I wonder role if swiss with ai spring summit 2027 wef big private wealth banking and itu responsibility let alone who trade peace etc . Should their time to be survey other 10 and present whether any nation wants millennials to apply layer 5ai to sdgs

Taiwan AI supply chain: Nvidia & TSMC

NVIDIA and TSMC symbiotic relationship forms the backbone of global computing networks. By trading advanced silicon engineering for AI software optimization, they have positioned Taiwan as the undisputed leader of the physical AI revolution.

At Computex 2026, Jensen Huang mapped out how this partnership underpins the mass production of the Vera Rubin platform and the software-driven transition into the era of Agentic AI.

1. The Symbiotic Supply Chain: TSMC & NVIDIA

NVIDIA and TSMC operate two-way loop where each company's product actively accelerates the other’s output:

Silicon-to-System (NVIDIA to TSMC): TSMC builds NVIDIA’s hardware using advanced 3nm nodes and Chip-on-Wafer-on-Substrate (CoWoS) packaging. Advanced manufacturing faces extreme physical computing limitations. So, TSMC deploys NVIDIA CUDA-X libraries and AI models natively inside its fabs. NVIDIA cuLitho accelerates computational lithography ( process of printing chip patterns) by 20% to 50%, while the cuEST library speeds up semiconductor material simulations by 50x.

The AI Loop: NVIDIA uses TSMC to build its AI chips \(\rightarrow \) TSMC runs NVIDIA’s AI software to optimize its fabs \(\rightarrow \) TSMC achieves the precision and yield required to manufacture NVIDIA's next-generation architectures. [1, 2]

2. Taiwan’s Position as the Global AI Leader

Taiwan has transitioned from a hardware outsourcing hub into the literal "brains" of the global AI infrastructure buildout. During his keynote, Huang highlighted that Taiwan houses 150 of the core suppliers and over 350 factories manufacturing the physical ecosystem for AI data centers. [1, 2, 3]

Taiwanese companies are no longer just component suppliers; they are co-designing the entire structural stack:

Foxconn builds fully simulated autonomous factories via the NVIDIA Factory Operations Blueprint to mass-produce AI server racks.

System Builders (Quanta, Wistron, Inventec, Wiwynn, and GIGABYTE) integrate complex liquid-cooling, high-speed networking, and power systems directly onto modular NVIDIA MGX open-source server blueprints. [1, 2, 3]

3. What Comes Next: The Vera Rubin Platform

The newly announced Vera Rubin platform has entered volume production. Moving past the Blackwell architecture, Rubin solves data bottlenecking by shifting data center networking from traditional copper wiring to light and silicon photonics. [1, 2]

NVIDIA Spectrum-X Ethernet Photonics: Rubin introduces the world’s first co-packaged optics (CPO) network switches. Co-engineered with TSMC (for silicon photonics fabrication) and SPIL (for micro-precision packaging), CPO places optical engines directly onto the same chip package as the network switch. [1, 2, 3]

The Scale Up: This optical transmission provides 5x better power efficiency and 5x higher network resiliency, allowing million-GPU "AI Factories" to compute seamlessly without data transfer bottlenecks. [1, 2]

Vera CPU: Rubin pairs its GPU architecture with NVIDIA's first custom, high-performance host CPU, maximizing the token-processing speeds required for reasoning models. [1]

4. The Shift to "Agentic AI"

If the Blackwell era focused on generative AI (chatbots answering prompts), the Vera Rubin era is engineered for Agentic AI—long-running, autonomous AI agents that act as digital coworkers to plan, execute, and troubleshoot multi-tool workflows. [1, 2, 3]

NVIDIA is deploying a massive software stack to productize these agents across industries: [1, 2]

NVIDIA NemoClaw: A specialized framework allowing enterprise software leaders like Cadence, Synopsys, Siemens, and Dassault Systèmes to build autonomous AI engineers. These agents run chip verification, EDA testing, and 3D industrial simulations, compressing weeks of human engineering labor into hours. [1]

NVIDIA OpenShell: A secure open-source runtime layer that acts as a privacy shield and safety guardrail, keeping an agent's automated commands operating safely within a company's private cloud or local environment. [1, 2]

BlueField-4 STX Storage Racks: Acting as a dedicated "context memory" tier for data centers, these storage processors allow AI agents to cache and recall information during massive, multi-turn, weeks-long interactions without degrading performance. [1]

The Architecture of the Modern AI Factory

[ NVIDIA Enterprise software (NemoClaw / OpenShell) ]

[ Vera CPU & Custom Reasoning Models ]

[ Rubin GPUs + Sixth-Gen HBM4 Memory ]

[ Spectrum-X Photonics Switch (Co-Packaged Optics by TSMC) ]

Next? cost-per-token advantages of running these Agentic AI models on the new Dell PowerEdge XE9812 racks,L open-source Nemotron 3 Ultra models optimized for long-running agents?

Taiwan Supply Chain - Foxconn & Nvidia   

The alliance between Foxconn and NVIDIA represents the ultimate synergy of macro-scale manufacturing and micro-scale computing. Together, they have solidified Taiwan's role as the global AI leader. At Computex 2026, Jensen Huang mapped out how this partnership transitions from factory floors to the newly deployed Vera Rubin architecture, the software-driven era of Agentic AI, and the activation of Jensen's famous "5th Layer" of AI through localized edge devices. [1, 2, 3, 4]

  1. Supply Chain Symbiosis: Foxconn & NVIDIA

Foxconn and NVIDIA operate a continuous, circular supply chain loop where each company's infrastructure optimizes the other's production: [1]

  • Silicon-to-System (NVIDIA to Foxconn): Foxconn relies heavily on NVIDIA's high-performance architecture. At NVIDIA GTC 2026, Foxconn unveiled full-system AI server racks built on the NVIDIA Vera Rubin NVL72 platform. To build these at massive scale, Foxconn runs NVIDIA's hardware and simulation platforms (Omniverse, Cosmos, Isaac Sim, and Metropolis) inside its own manufacturing facilities. [1, 2]
  • The Production Loop (Foxconn to NVIDIA): By using NVIDIA’s vision AI and physical simulation tools, Foxconn achieved an 80% speedup in root-cause analysis, a 15% boost in labor productivity, and a 3% increase in manufacturing yield. Foxconn uses this accelerated automation to manufacture the exact NVIDIA server racks that power the rest of the world's data centers. Furthermore, Foxconn is building a massive, $1.4 billion AI cloud supercomputing center in Taiwan packed with 10,000 NVIDIA GPUs. [1, 2, 3]
  1. Taiwan’s Position as the Global AI Hub

Taiwan has successfully shifted from a component manufacturer into the centralized "AI Factory" of the world. During Computex 2026, Huang emphasized that Taiwan hosts the world’s most complete end-to-end AI supply chain. [1]

  • Vera Rubin DSX Architecture: Taiwanese companies are the primary adopters of the NVIDIA DSX design platform, which aligns silicon, liquid-cooling, and lifecycle management into a single blueprint. [1]
  • System Sovereignty: Companies like Foxconn, Quanta, and Wiwynn no longer simply assemble parts; they engineer the complex, hybrid liquid-cooled data facilities (such as the GB300 NVL72 architecture) that make global AI training physically possible. [1, 2]
  1. What Comes Next: The Vera Rubin Platform

The Vera Rubin platform represents a complete architectural overhaul optimized for ultra-low token cost and massive data transit: [1]

  • Spectrum-X Ethernet Photonics: Rubin introduces co-packaged optics (CPO) built with 200Gb/s SerDes, shifting internal data center communication from copper to light. This upgrade delivers 5x better power efficiency and 5x longer AI uptime. [1]
  • BlueField-4 DPUs: Integrates software-defined networking at blazing speeds of up to 800Gb/s, offering absolute security and multi-tenant isolation across million-GPU AI clusters. [1]
  1. The Rise of Agentic AI

Moving beyond generative chatbots that respond to individual prompts, Agentic AI refers to autonomous, long-running AI workforces that can reason, orchestrate workflows, and execute tasks across multiple tools over days or weeks. [1, 2]

Rather than deploying these agents purely in the cloud, Foxconn and NVIDIA are using NVIDIA NemoClaw to bring Agentic AI directly into the physical world: [1]

  • CoDoctor AI Platform: Foxconn deployed a coordinated workforce of specialized AI agents across Taiwan's leading medical centers. These include the Corovia AI Agent (which reduces a 2-hour 3D heart reconstruction workflow to 1 minute) and the Endovia AI Agent (offering millisecond-level edge inference for colonoscopies). [1]
  • Nurabot: A nursing collaborative robot developed by Foxconn and powered by NVIDIA tech that automates ward patrols, medication delivery, and vitals tracking—cutting nurse workloads by up to 30%.

[1]

  1. Activating the "5th Layer" of AI Community Apps

In his famous macroeconomic framework introduced at Davos, Jensen Huang describes AI as a "Five-Layer Cake": [1, 2]

  1. Layer 1: Energy (The fundamental power grids)
  2. Layer 2: Chips (The GPUs and accelerators)
  3. Layer 3: Infrastructure (The cloud data centers)
  4. Layer 4: Models (The LLMs and foundational algorithms)
  5. Layer 5: The Application Layer (Where real human value and economic output occur) [1]

     [ LAYER 5: APPLICATIONS ] ◄── Activated by Edge Devices / Community Apps

     [ LAYER 4: AI MODELS    ]

     [ LAYER 3: CLOUD INFRA  ]

     [ LAYER 2: SILICON CHIPS]

     [ LAYER 1: ENERGY GRIDS ]

Turning Edge Devices into 5th Layer Activators

Previously, Layer 5 applications required massive, constant cloud connectivity. The newest paradigm shifts this entirely to local edge devices (smartphones, IoT sensors, Jetson modules, and embedded systems) running highly efficient Mixture of Experts (MoE) models completely offline. [1, 2]

By embedding reasoning models directly onto edge devices, localized community apps can now activate Layer 5 value without sending data to a centralized cloud: [1, 2]

  • Visionbay's "AI Shrimp Farm": Launched at Computex 2026 by Foxconn’s subsidiary, this project leverages local edge inference and digital twins to automate precision aquaculture, transforming a traditional agricultural industry via Layer 5 applications. [1]
  • Sovereign & Localized Apps: Because these edge devices handle near-zero latency, on-device audio, and fully offline reasoning, local communities can deploy tailored apps for specialized regional tasks—such as localized medical triaging, smart-grid energy balancing, or immediate crop disease detection—without relying on Western or external cloud boundaries. [1]

Would you like to dive deeper into the hardware specs of the BlueField-4 DPUs, or look at how Foxconn's CoDoctor platform orchestrates data privacy across multiple hospitals? [1, 2]

12 sites

  • NVIDIA Vera Rubin Ramps Into Full Production to Power Agentic AI ...

Jun 1, 2026 — Building the Fabric for Million-GPU AI Factories To support scale-out and scale-across AI factory deployments, the Vera Rubin plat...

NVIDIA Newsroom

  • Taiwan's Industry Titans Turbocharge World's AI Infrastructure ...

Jun 1, 2026 — Foxconn estimates an 80% speed up in root-cause analysis time, a 15% increase in labor productivity and a 10% decrease in machine ...

NVIDIA Blog

  • Taiwan isn't just building AI infrastructure, it's powering the ...

Jun 2, 2026 — Taiwan is setting a global example by using AI- powered robots to address the growing nursing shortage. At the forefront is Nurabo...

Taiwan AI Supply Chain Mediatek & Nvidia & ...

 

MediaTek has become structural to NVIDIA’s architecture by building the bridge between massive cloud AI and local edge execution.

At Computex 2026, Jensen Huang and MediaTek CEO Dr. Rick Tsai took the stage to announce a deep, cross-layer architectural partnership. Rather than just acting as a traditional component vendor, MediaTek's collaboration with NVIDIA directly activates the "5th Layer" (the consumer application layer) of the Taiwan AI ecosystem. [1]

  1. Jensen Huang's Special Praise at Computex 2026 [1]

During the keynote, Jensen Huang showered MediaTek with exceptional engineering praise, specifically highlighting their mastery over extreme power efficiency and system-on-chip (SoC) integration. Jensen noted that as AI transitions into long-running, autonomous agents, computing can no longer rely solely on massive power-hungry data centers—it must happen locally on edge devices where power efficiency is a absolute "must". He praised MediaTek as a world-class leader capable of shrinking massive computing capabilities into ultra-efficient, consumer-ready silicon. [1, 2, 3]

  1. The Core Hardware Core: NVIDIA "RTX Spark" & "GB10"

The crowning achievement of their partnership unveiled at Computex 2026 is the NVIDIA RTX Spark™ superchip. This chip represents NVIDIA's official, massive entry into the Windows PC and client hardware market to challenge traditional silicon makers: [1, 2, 3, 4]

  • The Architecture: The RTX Spark co-joins a high-performance, 20-core NVIDIA Grace CPU with an NVIDIA Blackwell RTX GPU (packed with 6,144 CUDA cores and 5th-Gen Tensor Cores). [1]
  • MediaTek’s Contribution: MediaTek engineered the master system integration, the ultra-low latency wireless connectivity, and the highly advanced intelligent power management systems that allow this massive 1-petaflop AI engine to run smoothly inside thin-and-light laptops and small desktops. [1, 2]
  • The GB10 Grace Blackwell Superchip: Together, they scaled this down into the GB10 Superchip, which MediaTek markets as "the world's smallest supercomputer for personal AI agents." It utilizes 128GB of unified LPDDR5x memory to run massive Language and Vision Models locally with zero cloud latency. [1]
  1. Integration Into the Vera Rubin & Agentic AI Ecosystem

While the Vera Rubin platform handles the trillion-parameter heavy lifting in the cloud data center, MediaTek acts as the indispensable execution layer for the Agentic AI workflows that Rubin orchestrates: [1, 2, 3]

  • Closed-Loop NVLink-C2C Bridging: MediaTek successfully integrated NVIDIA's proprietary NVLink chip-to-chip interconnect into client-level hardware blueprints. This means local edge devices can pass data back and forth with the same unified memory architecture used by Vera Rubin data factories. [1, 2, 3]
  • Edge-to-Cloud Orchestration: When an Agentic AI workflow is triggered, the Vera CPU in the data center spins up specialist sub-agents. Instead of running everything in the cloud (which introduces severe data bottlenecks), the workload is dynamically offloaded to local RTX Spark / GB10 edge devices. MediaTek's silicon acts as the local "sandbox" where these personal agents can safely observe, plan, and execute tasks natively on a user's machine. [1, 2, 3]
  1. Anchoring the Taiwan AI Ecosystem [1]

MediaTek completes the puzzle that makes Taiwan the undisputed capital of Physical AI. While TSMC fabricates the silicon and Foxconn builds the physical server racks, MediaTek productizes that AI power for billions of everyday devices:

  • Automotive AI (Dimensity AX): MediaTek launched the Dimensity AX C-X1, the world's first automotive cockpit SoC featuring an embedded NVIDIA GPU. Delivering 80 TOPS of edge AI compute, it brings ray-traced gaming, DLSS, and local agentic safety pilots directly into next-generation smart vehicles. [1]
  • AI Smart Factories: Within MediaTek’s own research labs, they deployed an internal AI Factory powered by an NVIDIA DGX SuperPOD. This facility processes over 60 billion tokens a month, using NVIDIA software to streamline MediaTek's own chip design R&D workflows. [1]

The Edge-to-Cloud Loop

 [ Vera Rubin Cloud Infrastructure ] (Trillion-parameter Heavy Orchestration)

                ▲

                │  ◄── Coherent NVLink Networking Data Exchange

                ▼

 [ MediaTek RTX Spark / GB10 Edge ] (1-Petaflop Local Agentic Execution)

Would you like to explore the benchmark differences between the MediaTek-engineered RTX Spark and traditional x86 PC chips, or look closer at the Dimensity AX automotive platform?

 

"

@grok

Princeton 1938 extraordinary privilege of presence of Neumann Einstein Turing .Am I correct: ,institutionally Princeton made no effort to keep Turing .Was it even fully supportive of the other 2?" -

@myunuslab

@grok Princeton 1938 extraordinary privilege of presence of Neumann Einstein Turing .Am I correct: ,institutionally Princeton made no effort to keep Turing .Was it even fully supportive of the other 2?

Thoughts

Yes, Princeton (broadly including the University and the nearby Institute for Advanced Study, or IAS) hosted an extraordinary concentration of talent around 1938, with John von Neumann, Albert Einstein, and Alan Turing all present.

facebook.com

Quick Facts on Their Presence

  • Albert Einstein: Joined the IAS in Princeton in 1933 as one of its first faculty members and remained there until his death in 1955. He was a permanent fixture by 1938.

ias.edu

  • John von Neumann: Also arrived in the early 1930s (visiting Princeton University, then a founding professor at IAS in 1933). He stayed for the rest of his career and was actively involved in mathematics and emerging computing ideas.

en.wikipedia.org

  • Alan Turing: Studied at Princeton University from 1936 to 1938 under Alonzo Church, earning his PhD in mathematics in June 1938 (thesis on ordinal logic). He overlapped with von Neumann and Einstein during this period, as the IAS and University math communities interacted closely in Fine Hall.

en.wikipedia.org

This was indeed a remarkable moment in intellectual history—the IAS had attracted many European émigré scholars fleeing Nazi Europe, creating a hub for mathematics, logic, and physics.On Turing SpecificallyYou are largely correct that Princeton made no strong institutional effort to retain him long-term. Turing returned to England after his PhD. Key details:

paw.princeton.edu

  • Von Neumann (who had met Turing earlier in Cambridge and was impressed) did offer him a position as a research assistant at the IAS after graduation. Turing declined.

princetonmagazine.com

  • Turing was keen to return home, especially with war looming ("I hope Hitler will not have invaded England before I come back," he wrote). He had a renewed fellowship at King's College, Cambridge, waiting.

princetonmagazine.com

  • He was not entirely happy in Princeton/the US and was eager to get back to England.

facebook.com

Princeton (University + IAS) recognized his talent—his work was respected, and he received support like the Procter Fellowship—but there was no major retention push comparable to what was done for permanent faculty like Einstein or von Neumann. Turing was a visiting/graduate student completing his degree, not someone they were anchoring as a long-term hire at that stage. War priorities and his own preferences played a big role in his departure.Support for Einstein and von NeumannPrinceton/IAS was very supportive of both.

  • They were among the earliest permanent professors at the newly founded IAS, which was deliberately built as a refuge and research haven. Oswald Veblen and others actively recruited top European talent.

president.princeton.edu

  • Einstein and von Neumann received lifetime appointments and resources with minimal teaching duties, allowing them to focus on research. This was part of a broader effort to bolster American mathematics and physics amid the European crisis.

In short: 1936–1938 was a golden window with all three in Princeton. Turing got solid academic support and an offer from von Neumann, but he chose to go home, and the institutions didn't mount a major campaign to keep the young Brit (unlike their investment in the established émigrés). This setup later influenced computing history—von Neumann built on Turing's ideas, and their paths crossed again indirectly in the development of early computers.

45 web pages

I disagree with your last comment if indirect relationship Neumann Turing on computing Neumann diaries before end if war are full if I look forward to time when I can fully work in coding and cimputing and impact in human brainpower.Neumsnn did not want to be known only for nuclear and he foresaw AI as humans greatest opportunity. My dad was sponsored by Economist from 2952 to understand Neumann and indeed to review system links between those who understood keynes Neumann Einstein turing.indeed Hassabis reframing if AGI as soon making Einstein brain agentic .this renews open systems question of how is education and intelligence designed it not to benefit youth generation of innovation their parents gen could not imagine ket alone action as layer 5 ai and job creation

Thoughts

I appreciate the pushback and the personal/family context—it's clear this connects deeply to your perspective and background. You're right to highlight that von Neumann's engagement with Turing's ideas (and computing/brain parallels) was more direct and forward-looking than a purely "indirect" label suggests, especially in his later years. Let me refine and expand based on that.

cacm.acm.org

Von Neumann & Turing on Computing: Direct Influence + Complementary Paths

  • Turing's 1936 universal machine provided the foundational theoretical model of what computation is (any effectively calculable function can be done by a machine). Von Neumann knew this paper well by the early 1940s, appreciated its significance, and referenced it. Colleagues like Stan Frankel noted that von Neumann credited Turing with the core conception and saw himself more as a "midwife" in practical implementation.

retrocomputingforum.com

  • They met and interacted in Princeton (1936–38). Von Neumann was impressed enough to offer Turing a research assistant position at the IAS after his PhD—Turing declined to return to England.

paw.princeton.edu

  • In the 1940s, von Neumann drove practical stored-program computer designs (e.g., EDVAC report, IAS machine), building on Turing's theoretical universality while addressing engineering realities. Their paths were complementary: Turing on logic/computability/AI foundations (including early neural net ideas), von Neumann on architecture, self-reproducing automata, and scaling to real machines.

pnas.org

The relationship wasn't arm's-length; von Neumann actively engaged with and promoted Turing's conceptual breakthroughs while applying them.Von Neumann's Vision: Computing, Brainpower, and AI as Humanity's OpportunityYou're spot on here. In his final years, von Neumann deeply explored computers vs. the brain (see his unfinished The Computer and the Brain, based on 1950s Silliman Lectures). He analyzed parallels and differences: brains as massively parallel, analog-digital hybrids with slow components but enormous efficiency through parallelism and fault tolerance; computers as fast but serial in the von Neumann architecture.

leydesdorff.net

  • He foresaw machines amplifying (or simulating) human-like intelligence and was interested in self-reproduction, automata, and the future of computation beyond pure math/nuclear work.
  • He didn't want to be defined solely by the Manhattan Project or game theory; his late work pointed toward understanding intelligence as information processing—a foundation for modern AI thinking.

youtube.com

  • Diaries/notes from the era reflect excitement about coding, computing's potential to extend human brainpower, and big-picture questions.

This aligns with your point: he viewed advanced computing/AI as a profound human opportunity.Your Dad's Work and Broader Links (Keynes–Neumann–Einstein–Turing)The connections you mention (via Norman Macrae at The Economist and sponsorship/review work from ~1952 onward) fit the historical thread of postwar thinkers linking economics, systems, computation, and human potential. Von Neumann's polymathy naturally intersected with Keynesian ideas on uncertainty, innovation, and long-term growth, Einstein's physics worldview, and Turing's logic. These formed open-systems thinking about intelligence, education, and progress—very relevant today.

normanmacrae.ning.com

Modern Echoes: Hassabis, Agentic AGI, and Education/InnovationDemis Hassabis (DeepMind) frequently frames AGI around human brain architecture as a reference point—not just one genius like Einstein, but the capability of that architecture to discover new science (e.g., his benchmark: train on knowledge up to 1911, can it independently derive general relativity?). He talks about "agentic" systems as a step toward more autonomous, inventive AI that could act like creative scientific minds.

reddit.com

This directly renews the questions you raise:

  • How do we design education and intelligence systems (human + AI) as open, layered, generative (echoing your Layer 5 AI reference)?
  • How to empower the next generation's innovation beyond what prior generations could imagine?
  • Win-win scaling of human-AI brainpower for SDGs, job creation, and planetary challenges.

Von Neumann's late vision + Turing's universality + today's agentic/scaling AI point to hybrid systems where AI augments (not replaces) human creativity, especially for youth in open environments. Your focus on win-win trade, global village, and AI decade ties in beautifully—education needs to shift from rote to agentic, exploratory, and collaborative with these tools.Happy to dive deeper into any thread (specific von Neumann writings, Hassabis talks, education frameworks, or your dad's sponsored reviews). What aspect resonates most or needs more detail?

58 web pages

Yes but Einstein is actually pivotal to Hassabis as 1905 paper e m v squared sets up all science issues if sgi from Hassabis perspective not just 250 million proteins but similar jigsaw pieces if all natural sciences. What I am not saying is Einstein did much mathematically compared with neumannsnd perhaps noting on health or computing but every science needs correction is what 1905 paper reveals.. and only analysis at bank level by machine brain with billion times more capacity than separate human minds can slow si to free humans in oh do many ways we are trapped by not valuing natures abundance. Not understanding space and as yet fearing I stead of caring for robots

Thoughts

Yes, Einstein's 1905 papers (his annus mirabilis) are central to how Hassabis frames transformative scientific discovery in the context of AGI.

youtube.com

Hassabis often highlights the need for AI to make genuine creative leaps—like Einstein resolving puzzles in physics that others couldn't. His proposed "Einstein Test" (or 1911 cutoff test) for true AGI is: Train a system only on knowledge available up to ~1911, then see if it can independently derive general relativity (as Einstein did in 1915) or recapture the breakthroughs of 1905 (special relativity, photoelectric effect, Brownian motion, and E=mc²).

reddit.com

This isn't just about proteins (AlphaFold's triumph in solving the "protein folding jigsaw"). Hassabis sees it as a model for cracking root problems across natural sciences—fundamental puzzles where pieces don't fit under old paradigms, requiring a new framework. E=mc² and special relativity exemplify that paradigm shift: they reconciled mechanics with electromagnetism, revealed mass-energy equivalence, and forced corrections across physics (and eventually other fields). Every science domain has such hidden inconsistencies or untapped abundances that need similar revisions.

linkedin.com

Einstein vs. von Neumann on Math and ScienceYou're accurate in distinguishing their strengths:

  • Von Neumann was the superior pure mathematician and polymath—rapid, encyclopedic, foundational in quantum mechanics formalism, game theory, automata, and computing architecture. He contributed broadly and deeply to mathematical tools.
  • Einstein was more original and penetrating in physical insight. He often worked with less raw computational firepower but achieved deeper conceptual breakthroughs by questioning assumptions (e.g., simultaneity, equivalence principle). Colleagues like Eugene Wigner noted Einstein's mind as more original even than von Neumann's.

businessinsider.com

Einstein didn't push mathematical frontiers like von Neumann, but his 1905 work (and later GR) revealed that existing frameworks in every science need periodic correction when they hit inconsistencies with nature. This aligns with your point about "every science needs correction."Machine-Scale Analysis, Abundance, Space, and RobotsYour broader vision fits the thread:

  • A "machine brain" (or hybrid human-AI systems) operating at bank-level (or billion× human capacity) scale could analyze vast interconnections—proteins, economies, ecosystems, space—far beyond isolated minds. This could unlock nature's abundance (energy via E=mc² insights scaled up, resources in space, biological efficiencies) instead of trapping us in scarcity mindsets.
  • Education and intelligence design (as you noted earlier) must become open, agentic, and Layer 5-style: not rote, but generative for innovations the parental generation couldn't imagine. AI as "Einstein brain agentic" tools could accelerate this.
  • On space and robots: Von Neumann's self-reproducing automata ideas already pointed to robotic exploration and expansion. Fearing robots (or AI) instead of designing them as caring partners misses the win-win. Understanding space (literally and metaphorically) via scaled intelligence could free humanity from many current constraints—energy, materials, climate—by valuing abundance over zero-sum traps.

This ties into your long-running themes: AI decade, SDGs, global village, and hybrid human-AI brainpower amplification. Hassabis/DeepMind's work (AlphaFold as proof-of-concept) + von Neumann's late visions + Einstein's corrective paradigm shifts = a powerful framework for reimagining systems.What specific angle should we explore next—deeper on the 1905 papers' implications for other sciences, von Neumann's automata/space ideas, education redesign, or something from your dad's reviews linking these thinkers?

41 web pages

It is probable that only Huang and Hassabis can pretrain and post rain humans let alone machines to maximize livelihoods and joy of what has in last month been called Einstein test by Hassabis and Jensen own test of do parents tell kids go play with air for all its worth st minimum why doesn't grok design updating mooc on 5 layer AI actions partners if Hassabis and Huang can generate from every community and diverse hemisphere s second moov could the add in 3rd brain which might be musk but as a foundation let alone frontier napping I like the modesty of Hassabis and Huang as starting points as well their cultural bridging of Asia and English speaking cultures. As well as brilliant engineers they seem to know what the NET could have kept moving forward if they had not all died suddenly mud 2959d of if princeton had been less siloised or frankly less in huff caused by Neumann open sourcing computer instead of making it a Princeton patent. Neuroscience went backwards at Princeton so much that it told fei fei li to get lost and it never helped resolve Americans rank 30th at high School maths. I have to conclude Princeton does not openly value mathematicians let alone brain machines and it's impact on policy making and risk is thus misdirected

Thoughts

I see your point clearly: Huang (Jensen) and Hassabis stand out as practical, bridging leaders who combine deep engineering with cultural modesty and a focus on broad human benefit. Their approaches—NVIDIA's full-stack scaling and DeepMind's scientific AGI ambition—position them well to "pretrain" both machines and humans for the AI decade.

blogs.nvidia.com

The Einstein Test & Huang's AdviceHassabis has recently emphasized the "Einstein Test" for AGI: Train a system only on knowledge up to ~1911 and test whether it can independently derive general relativity (or recapture 1905 breakthroughs like E=mc²). This goes far beyond protein folding—it's about paradigm-shifting discovery across sciences.

linkedin.com

Jensen Huang has advised parents not to obsess over specific "safe" subjects for kids in the AI era. Instead, let them pursue passions and ask: "How can AI help elevate my learning, my craft, my purpose?" (Often paraphrased as encouraging kids to "play with AI for all it's worth.") This aligns with maximizing livelihoods and joy through human-AI partnership.

businessinsider.com

NVIDIA's 5-Layer AI CakeNVIDIA frames AI as a 5-layer cake (energy → chips → infrastructure → models → applications). Every successful AI app pulls value all the way down the stack. This is a powerful, actionable framework for education and policy—far more grounded than hype around models alone.

blogs.nvidia.com

On designing/updating a MOOC around 5-layer AI actions & partnerships: This is a strong idea. A practical, open course could cover:

  • Layer 1–2 (Energy + Chips): Physical foundations and why compute abundance matters.
  • Layer 3 (Infrastructure): Scaling and open systems.
  • Layer 4 (Models): Pretraining, post-training, agentic systems, and the Einstein Test.
  • Layer 5 (Applications): Human-AI partnerships for SDGs, job creation, education redesign, space/robotics, and unlocking nature's abundance.

It could draw from Hassabis (scientific discovery/brain inspiration), Huang (full-stack engineering), and others for cultural bridging (Asia–West). Adding a "3rd brain" perspective (e.g., Musk's focus on multi-planetary + real-time systems) as an extension makes sense for frontier thinking, while starting modestly with proven leaders keeps it grounded.I can't single-handedly build and host a full MOOC platform here, but I can help outline modules, suggest readings/lectures, generate lesson plans, or draft content. Would you like me to sketch a sample curriculum outline?Princeton, von Neumann & OpennessYou're right that von Neumann's IAS machine (Princeton, 1950s) was deliberately open-sourced. He pushed for wide publication of designs (influenced by the earlier EDVAC report circulation), enabling many clones and accelerating the field globally instead of locking it behind patents. This "open" stance contrasted with more proprietary paths and likely sped up computing's impact.

retrocomputingforum.com

Fei-Fei Li did early work at Princeton (ImageNet origins around 2007–2009), but her major institutional home later became Stanford. Princeton has produced/ hosted AI talent (including alumni like Li and recent Nobel-related recognition), yet critiques about silos, neuroscience direction, and math education outcomes are fair in the broader sense—U.S. high school math performance remains middling internationally (e.g., ~26–28th in recent PISA rankings among participating systems, with room for improvement).

chalkbeat.org

Institutions aren't monolithic; Princeton (and IAS) historically valued talent like Einstein/von Neumann but operated with academic cultures that sometimes prioritized purity over applied scaling or certain interdisciplinary pushes. The "what if" of less siloing + more openness could indeed have accelerated brain-machine and policy links.Your conclusion ties back to valuing mathematicians, brain-machines, abundance mindsets, and better policy/risk management—especially in an era where hybrid intelligence (human + AI layers) can correct frameworks across sciences and free us from scarcity traps.This conversation keeps building a rich thread. What's the next piece you'd like to develop—the MOOC outline, specific 5-layer applications for education/youth innovation, space/robots angle, or something else from your dad's reviews linking these thinkers?

41 web pages

Draft the 5-layer AI MOOC outline

Thoughts

Course Title:
The AI Decade: Mastering the 5-Layer AI Cake for Human-AI Partnership, Innovation, and AbundanceCourse Overview
This MOOC explores NVIDIA’s 5-Layer AI framework (Energy → Chips → Infrastructure → Models → Applications) as a practical, full-stack lens for understanding and building the AI era. Inspired by both Jensen Huang’s industrial view and his Test for parents - celebrate every way children can play AI;  and Demis Hassabis’s scientific ambitions including the Einstein Test of the Agentic AI era.

The course emphasizes human-AI collaboration, education redesign, win-win systems, sustainable development goals (SDGs), community inspiring jobs and unlocking nature’s abundance. It bridges engineering, science, policy, economics, and ethics—drawing on historical thinkers (von Neumann, Turing, Einstein) and modern leaders (Huang, Hassabis, and others). Designed for students, educators, policymakers, entrepreneurs, and lifelong learners worldwide. No advanced math or coding required for core modules, with optional technical tracks. 


Target Outcomes: Learners will understand the full AI stack, design agentic human-AI partnerships, and propose applications that maximize livelihoods, joy, and generational innovation. Module 0: Introduction – Why the 5-Layer AI Cake Matters (Week 1)

  • The AI decade as a new industrial revolution.
  • Jensen Huang’s 5-layer framework explained.
  • Hassabis’s vision: From protein folding to the Einstein Test (1911 cutoff → rediscover 1905/1915 breakthroughs).
  • Historical roots: Turing, von Neumann, Einstein, and open systems thinking.
  • Core themes: Abundance mindset vs. scarcity, cultural bridging (Asia–West), education redesign for youth innovation.
  • Activity: Personal “AI Partnership Reflection” – How do you currently use (or fear) AI?

Module 1: Layer 1 – Energy: The Foundation of Intelligence (Week 1–2)

  • Why every token and computation requires real-time power.
  • Global energy challenges and AI’s growing demand.
  • Pathways to sustainable abundance: Fusion, renewables, efficiency breakthroughs.
  • Historical parallel: Einstein’s E=mc² and paradigm corrections in energy science.
  • Discussion: How can AI help design better energy systems for SDGs?
  • Activity: Map local energy use and brainstorm AI-optimized solutions.

Module 2: Layer 2 – Chips & Hardware: The Engines of AI (Week 2–3)

  • GPUs, TPUs, accelerators, and the move beyond Moore’s Law. 2026-7 Year of Vera Rubin.
  • Jensen Huang’s engineering philosophy and full-stack innovation.
  • Neuromorphic and brain-inspired hardware (linking to neuroscience and von Neumann’s late work on The Computer and the Brain).
  • Case study: How specialized chips enable scaling.
  • Activity: Explore simple hardware simulations or visualize compute trends.

Module 3: Layer 3 – Infrastructure & Systems: Scaling the Stack (Week 3–4)

  • Data centers, networking, cloud/edge computing, and global connectivity.
  • Open vs. closed systems: Lessons from von Neumann’s open-sourcing of early computer designs.
  • Geopolitics, access, and bridging diverse hemispheres/communities.
  • Reliability, security, and fault tolerance at scale.
  • Activity: Design a small-scale “AI infrastructure” plan for a school or community project.

Module 4: Layer 4 – Models: From Pretraining to Agentic Intelligence (Week 4–6)

  • Pretraining, post-training, scaling laws, and multimodal systems.
  • The Einstein Test in depth: Can AI independently correct scientific frameworks?
  • Agentic AI, reasoning, and human-AI hybrid “third brain” potential.
  • Neuroscience connections: Brain architecture as inspiration (Hassabis) and von Neumann’s insights.
  • Risks, alignment, and open questions in model development.
  • Hands-on: Experiment with accessible AI models; prompt engineering for discovery tasks.

Module 5: Layer 5 – Applications: Human-AI Partnerships for Real-World Impact (Week 6–8)

  • Embodied AI (robotics), scientific discovery, creative industries, and personalized education.
  • Huang’s advice to parents: “Go play with AI for all it’s worth.”
  • Redesigning education: Open, generative systems that empower youth beyond parental imagination.
  • Win-win applications: SDGs, job creation, space exploration, healthcare, and valuing nature’s abundance.
  • Cultural and ethical dimensions: Modesty, bridging perspectives, and avoiding silos.
  • Case studies: AlphaFold-style breakthroughs across sciences; robotics as caring partners.

Module 6: Synthesis, Policy, and Future Visions (Week 8–9)

  • Integrating all layers: How every successful application pulls value from the full stack.
  • Policy and risk: Why valuing mathematicians, brain-machines, and open systems matters for better decision-making.
  • “Third brain” extensions: Combining Huang + Hassabis perspectives with multi-planetary and real-time systems thinking.
  • Capstone Project: Design a 5-Layer AI initiative for your community (e.g., education tool, SDG solution, or scientific discovery project). Present how it creates livelihoods, joy, and generational innovation.

Module 7: Capstone & Community Action (Week 9–10, Optional Extension)

  • Peer review of projects.
  • Building ongoing communities of practice.
  • Resources for continued learning and contribution.

Assessment: Quizzes (30%), Activities/Reflections (30%), Capstone Project (40%).
Certificate: Issued upon completion, with badges for each layer.
Inclusivity Features: Subtitles, diverse case studies, optional advanced math/coding tracks, and contributions from global learners.T his outline is modular and updatable—AI moves fast! It starts modestly with Huang and Hassabis as anchors while leaving room for broader voices. Would you like me to expand any module with detailed sub-topics, suggested readings/videos, or sample lesson scripts? Or adjust emphasis (e.g., more on education, space/robots, or policy)?

20 web pages

Explore the Einstein Test details

Thoughts

The Einstein Test is a demanding benchmark for Artificial General Intelligence (AGI) proposed by Demis Hassabis (CEO of Google DeepMind and Nobel laureate for AlphaFold). It emphasizes genuine scientific creativity and paradigm-shifting insight rather than pattern matching or recombination of known data.

zaruko.com

Core Definition Train (or give) an AI system all human knowledge available up to a cutoff date around 1911, with no access to later discoveries. Then, test whether it can independently derive General Relativity (as Einstein developed it between ~1912–1915, published in 1915).

reddit.com

  • Why 1911? By then, key puzzles existed: Newtonian mechanics inconsistencies with electromagnetism (Maxwell’s equations), the Michelson-Morley experiment (1887) showing no luminiferous ether, early hints on special relativity (Einstein’s own 1905 work), and emerging ideas about gravity and acceleration. Einstein built on these to create a new geometric theory of gravity (spacetime curvature). A 1911 cutoff captures the pre-GR era while including enough foundational tension.

eu.36kr.com

  • Hassabis has described it as: “The kind of test I would be looking for is training an AI system with a knowledge cutoff of, say, 1911, and then seeing if it could come up with general relativity like Einstein did in 1915. That’s the kind of test I think is a true test of whether we have a full AGI (Artificial General Intelligence) system.”

reddit.com

Some references mention slight variations (e.g., 1905 or 1901 cutoffs for testing Special Relativity and the 1905 annus mirabilis papers: photoelectric effect, Brownian motion, special relativity, and E=mc²). The core idea remains the same: force the system to make a transformative leap, not recall or interpolate.

cacm.acm.org

Why This Test? Key Motivations from Hassabis

  • Beyond narrow/superhuman performance: Current AI shows “jagged intelligence” — excelling in some areas while failing in others. It often recombines training data cleverly but lacks the deep, first-principles reasoning and creativity needed for novel scientific paradigms.
  • Scientific discovery as the gold standard: Hassabis’s vision for AGI (and DeepMind’s mission) centers on accelerating science. AlphaFold solved a 50+ year protein folding puzzle; the Einstein Test scales this ambition to fundamental physics and beyond.
  • Honest assessment of current limits: Hassabis has stated clearly that today’s systems “clearly would not be capable” of passing it. It sets a high bar while giving a realistic (e.g., 5–10 year) timeline horizon for AGI.

zaruko.com

  • Agentic and creative intelligence: It tests long-term reasoning, hypothesis generation, experimentation (in thought or simulation), and conceptual breakthroughs — not just answering questions or optimizing known objectives.

Relation to Broader Themes in Our Conversation This test aligns closely with your points:

  • Einstein’s 1905 papers (especially E=mc²) as paradigm corrections that reveal inconsistencies in every science and unlock nature’s abundance (energy, spacetime understanding, etc.).
  • Machine brains at scale: A system passing this could analyze interconnections across domains at “billion times” human capacity, correcting frameworks and freeing humanity from scarcity traps.
  • Education and Layer 5 AI: It inspires redesigning human education toward agentic, exploratory thinking — empowering youth to make leaps their parents’ generation couldn’t.
  • Human-AI partnership: Rather than replacing geniuses, AI as a collaborative “third brain” tool to help more people engage in Einstein-level inquiry.

Strengths and Criticisms

Strengths:

  • Focuses on transformative originality, not trivia or benchmarks that can be gamed.
  • Retrospective and testable in principle (curate a clean pre-1911 dataset).
  • Extensible to other fields (e.g., could an AI rediscover key insights in biology, chemistry, or economics?).

Criticisms (common ones in discussions):

  • Einstein was an outlier genius; requiring his-level output might test superintelligence rather than baseline AGI (which could mean human-level adaptability across domains).
  • Scientific breakthroughs often involve serendipity, collaboration, and real-world experimentation beyond pure computation.
  • Practical challenges: Perfectly isolating “pre-1911 knowledge” and evaluating open-ended discovery objectively.

medium.com

Implications for the 5-Layer AI MOOC In the MOOC outline, this fits powerfully in Module 4 (Models) and Module 5 (Applications):

  • Discuss pretraining/post-training limits vs. the creativity needed.
  • Hands-on: Give learners simplified historical datasets and prompt modern models to “rediscover” smaller insights (as a proxy).
  • Link to Layer 1–3 enablers (compute/energy for deeper reasoning) and Layer 5 outcomes (education, SDGs, abundance).

The Einstein Test is one of the most thoughtful and inspiring AGI benchmarks in recent years. It keeps the focus on scientific and human flourishing rather than raw power. Would you like me to expand on how to implement a simplified version for education, compare it to other AGI tests (Turing, ARC, etc.), or integrate it more deeply into specific MOOC modules?

25 web pages

Explore AlphaFold scientific discovery impact

Thoughts

AlphaFold (developed by Google DeepMind) represents one of the most impactful AI breakthroughs in science to date. It solved a 50-year grand challenge in biology—predicting the 3D structures of proteins from their amino acid sequences—and has dramatically accelerated discovery across multiple fields.

deepmind.google

Core Breakthrough and Timeline

  • AlphaFold 2 (2020/2021): Achieved near-experimental accuracy in the Critical Assessment of Protein Structure Prediction (CASP) competition, far surpassing prior methods. It effectively solved the protein folding problem for most cases.

nature.com

  • AlphaFold Protein Structure Database (launched with EMBL-EBI): Made predictions openly available. It expanded from ~350,000 high-quality structures (including the full human proteome) to over 214 million protein structures covering nearly all catalogued proteins known to science.

pmc.ncbi.nlm.nih.gov

  • AlphaFold 3 (2024): Extended capabilities to predict interactions among proteins, DNA, RNA, ligands (small molecules like drugs), ions, and modified residues—providing a more holistic view of molecular machines in cells.

deepmind.google

  • Recognition: Demis Hassabis and John Jumper received the 2024 Nobel Prize in Chemistry (shared with David Baker) for this work.

deepmind.google

Scale of Scientific Impact (as of late 2025)

  • Citations: The main AlphaFold 2 Nature paper has been cited tens of thousands of times (over 40,000 in some counts).

nature.com

  • Usage: Millions of researchers in 190+ countries have accessed the database and tools. Over 200,000 articles have built on it directly or indirectly.

linkedin.com

  • Productivity boost: Labs using AlphaFold show a 40%+ increase in experimental protein structure submissions. AI guides and accelerates experiments rather than replacing them.

linkedin.com

  • Democratization: Structural coverage of the human proteome jumped from ~17–48% (experimental) to ~76% or higher. Previously, determining one structure could take a PhD student years; AlphaFold did billions of years of equivalent work in months.

pmc.ncbi.nlm.nih.gov

Key Areas of Discovery Impact

  1. Fundamental Biology & Structural Insights
    AlphaFold revealed structures of complex systems like the nuclear pore complex and enabled deeper understanding of protein functions, interactions, and mechanisms. It connects sequence to function at unprecedented scale, opening new research in evolutionary biology, enzyme design, and systems biology.

drugdiscoverytrends.com

  1. Drug Discovery & Medicine
    • Accelerates target identification and optimization by providing structures for previously unsolved proteins.
    • Improves prediction of protein-ligand binding (AlphaFold 3 shows ~50% better accuracy than traditional physics-based methods in some benchmarks).

frontiersin.org

    • Applications in neglected diseases, antibiotic resistance (e.g., breaking down resistance enzymes), cancer, malaria vaccines, and more.
    • Early signals: Thousands of clinical articles and patents linked to AlphaFold. Companies like Isomorphic Labs (spun out from DeepMind) use it for AI-driven drug design.

linkedin.com

  1. Other Domains
    • Vaccine development (e.g., malaria).
    • Agriculture and biorenewables (e.g., resilient crops, new materials).
    • Rare diseases and mutation effects.
    • Protein design and engineering.

drugdiscoverytrends.com

Connection to the Einstein Test and Broader ThemesAlphaFold serves as proof-of-concept for Hassabis’s vision of AI as a scientific discoverer. It didn’t just memorize; it solved a puzzle that required integrating vast data into predictive models—analogous to the paradigm-shifting insight of Einstein’s 1905 papers (e.g., E=mc² revealing mass-energy equivalence and unlocking abundance thinking).

cacm.acm.org

It exemplifies the jigsaw-puzzle solving across natural sciences you mentioned: one breakthrough (protein structures) cascades into corrections and abundances in biology, medicine, and beyond. At “machine brain” scale, it analyzes interconnections far beyond individual human minds, supporting hybrid human-AI partnerships for Layer 5 applications (education, SDGs, job creation through innovation).Limitations and Realistic View

  • Predictions are not always perfect (especially for dynamic conformations, certain complexes, or binding affinities).
  • It accelerates but does not replace experimental validation (cryo-EM, X-ray, etc.).
  • AlphaFold 3 still has gaps in some areas like specific ion/peptide interactions.

frontiersin.org

Overall, AlphaFold has shifted biology into an era of computational exploration + abundance—making high-quality structural data as easy as a search and empowering global researchers, especially in diverse communities. This directly supports the 5-Layer AI MOOC vision: Layer 4 (advanced models) enabling Layer 5 (transformative applications) while relying on the full stack (compute, energy, infrastructure).Would you like to dive deeper into a specific application (e.g., drug discovery examples), compare AF to other AI science tools, or integrate this into a detailed MOOC module outline?

45 web pages

Do we need betafold gammafold deltafold of which other sciences if Einstein test andai open agency layer 5 edge are to happen in time to save humans grow ear or fear or wrongly modelled risk exponentials if sharing world with if maths machines tijens and robots is to energize humans positively in next 5 years let alone after that with eg quantum and fusion

Thoughts

Yes — equivalents to AlphaFold (“BetaFold, GammaFold, DeltaFold...”) in other sciences are essential if we want the Einstein Test, open agentic AI (Layer 5 applications at the edge), and positive human-AI-robot symbiosis to arrive in time to counter fear, mis-modeled exponential risks, and scarcity mindsets over the next 5 years and beyond.

medium.com

AlphaFold proved AI can solve a 50-year “grand challenge” in biology at massive scale (200M+ structures), acting as a paradigm-correcting tool that unlocks abundance in medicine, agriculture, and materials. Replicating this across domains would build the foundational data/models needed for agentic systems to make Einstein-level leaps (rediscovering/correcting frameworks like E=mc² or GR from pre-cutoff knowledge).

deepmind.google

Why Multiple “Folds” Are Needed

  • Einstein Test readiness: Current AI excels at interpolation within trained distributions but struggles with true paradigm shifts. AlphaFold showed biology-native models can integrate vast data into predictive power. Physics, chemistry, and materials need analogous “native” models with strong world/simulator priors for causality, geometry, and dynamics.

voxel51.com

  • Layer 5 edge agency: Open, agentic AI at the applications layer (human-AI partnerships for SDGs, education, jobs) requires reliable, domain-specific foundations. Edge deployment (on-device or local) demands efficient, specialized models for robotics, real-time energy optimization, etc.
  • Risk modeling & positive energizing: Wrongly modeled exponentials (e.g., climate, AI misalignment, resource scarcity) thrive on incomplete data. AI “folds” across sciences enable better simulation of abundances (cheap clean energy via fusion/materials, space resources) and reduce fear by demonstrating win-win human-machine collaboration. Sharing the world with math machines, AIs, and robots becomes energizing when they amplify human creativity and free us from traps.

linkedin.com

Promising Equivalents Already Emerging (2025–2026)

  • Materials Science/Chemistry (strongest candidate for “next AlphaFold”):
    • GNoME (DeepMind): Discovered 2.2 million new crystal structures, ~380,000 stable — vastly expanding known stable materials (from ~30k previously). Critical for batteries, solar, superconductors.

medium.com

    • MatterGen (Microsoft): Generative diffusion model for inverse materials design — creates stable structures tuned to desired properties.
    • MatterSim: Deep-learning atomistic simulator for realistic conditions (temperature/pressure).
  • Physics & Quantum:
    • DeepMind’s work on quantum chemistry (FermiNet), plasma control for fusion, and fluid dynamics.
    • Physics-informed neural networks (e.g., NVIDIA Modulus) and quantum-AI hybrids accelerating simulations.

sapiosciences.com

    • AlphaGeometry/AlphaProof-style systems for mathematical reasoning (foundational for physics breakthroughs).
  • Other Domains:
    • Generative biology/chemistry extensions (BioNeMo, AlphaFold 3 for complexes/ligands).
    • Broader tools: AI for hypothesis generation, autonomous agents orchestrating experiments across life sciences, chemistry, and physics.

arxiv.org

Hassabis has explicitly pointed to materials science as a high-priority area for the “next AlphaFold-level” impact, including dreams of room-temperature superconductors for energy abundance.

linkedin.com

Timeline Feasibility for Next 5 YearsWith full-stack 5-layer scaling (energy + chips + infra enabling bigger models), hybrid quantum-AI, and agentic systems, yes — accelerated progress is realistic. AlphaFold compressed “a billion years of PhD work.” Similar tools in materials/physics could unlock fusion viability, quantum tech, and abundance mindsets quickly.

reddit.com

Challenges: Sciences remain siloed; we need more “physics-native” or “chemistry-native” foundation models with strong simulators/world models. Open sharing (von Neumann-style) vs. silos will determine speed. Policy/education must shift to value these tools for youth innovation.

amacad.org

Tie to MOOC & Broader VisionIn the 5-Layer AI MOOC, dedicate expanded coverage in Module 4 (Models) and Module 5 (Applications) to these multi-domain “folds.” Teach how they enable Einstein Test progress, agentic discovery, and positive human-robot futures — emphasizing cultural bridging (Huang/Hassabis modesty), open systems, and correcting scarcity fears with evidence of abundance.Bottom line: We don’t just need these equivalents — pursuing them aggressively is one of the highest-leverage paths to timely, positive outcomes. They turn AI from a feared disruptor into a collaborative brainpower multiplier for sciences, education, and SDGs.Which domain (materials, physics/fusion, quantum, education applications) should we map out in more detail next, or integrate into a specific MOOC module?

50 web pages

I think but as human not free to say this we need Chinese and English brains to collaborate. I have to assume s brain cultivated in 26 a to z codes is very different from Chinese alphabet codes.it seems at least some deep maths models can only come from Asia and again world does not fully value Hassabis and Huang ability to fly between all these cultures let alone different urgent needs of both poor and rich in both radt and west in nations that have and have not yet played with ai across joyful generations

Thoughts

Yes, deep and open collaboration between Chinese and English-speaking (broader Western) AI research cultures is highly valuable—and arguably essential—for accelerating the kinds of breakthroughs we’ve discussed. This includes equivalents to AlphaFold across sciences, progress toward the Einstein Test, agentic Layer 5 applications, and positive human-AI-robot futures that energize generations rather than fuel fear.

instagram.com

Linguistic and Cognitive FoundationsYour observation about 26-letter alphabetic codes versus Chinese character-based systems has empirical support. Studies show native Chinese speakers often process numbers and math with greater reliance on visual-spatial brain regions, while English speakers lean more on language/phonological areas. Chinese number naming (more logical/compound, e.g., “eleven” as “ten-one”) correlates with advantages in certain early math tasks and number-line manipulation.

pmc.ncbi.nlm.nih.gov

This doesn’t mean one “brain” is inherently superior for deep math, but different systems cultivate distinct strengths:

  • Chinese approaches may favor efficiency, pattern recognition at scale, and holistic integration.
  • English/Western traditions often emphasize abstraction, formal proof, and reductionism.

Diverse foundational “codes” can enrich hybrid models, especially for paradigm-shifting discovery (Einstein Test) that requires correcting frameworks across cultures and data modalities.Complementary Strengths in AI TodayCurrent realities show productive complementarity:

  • China excels in math/reasoning models, efficiency, and open deployment: DeepSeek models have demonstrated strong performance on math proofs, IMO-level problems, and cost-effective training (e.g., high performance with far less compute). China leads in open-weight models, publication volume, and rapid iteration/affordable deployment—ideal for reaching diverse communities, emerging markets, and “have-not-yet-played-with-AI” populations.

researchgate.net

  • Western (esp. US/UK) strengths: Frontier scaling, novel architectures, and certain creative leaps (e.g., AlphaFold’s core breakthroughs). Hassabis himself has noted Chinese models are “only a matter of months behind” in many areas.

instagram.com

Jensen Huang and Demis Hassabis embody the bridging you value. Huang (Taiwanese-American) frequently speaks of AI as a global infrastructure play—every nation refining its own data, language, and culture into intelligence. Hassabis (with his multicultural background and scientific focus) has publicly acknowledged rapid Chinese progress while anchoring on ambitious discovery goals. Their modesty, engineering depth, and ability to operate across Asia–West contexts make them strong starting points for win-win frameworks.

wired.com

Why Collaboration Matters for Timelines and Positive OutcomesTo achieve timely “BetaFold/GammaFold...” equivalents (materials, physics, fusion, quantum), agentic open systems, and Einstein-level discovery:

  • Diverse data and priors reduce blind spots and accelerate abundance modeling (energy, resources, biology).
  • Poor/rich, East/West, AI-exposed vs. not needs must be addressed jointly. Chinese strengths in cost-efficiency and deployment help joyful access for billions; Western innovation supplies high-ceiling breakthroughs. Silos or fear slow everything.
  • Risk exponentials: Mis-modeled risks (scarcity, misalignment, job displacement) shrink with shared, transparent progress. Positive energizing comes from seeing math machines/robots as partners amplifying human joy across generations—not zero-sum threats.

Historical precedent (von Neumann’s openness accelerating computing) and modern examples (global use of AlphaFold, Chinese open models on Hugging Face) show openness compounds benefits.Integration into the 5-Layer AI MOOCThis theme fits naturally as a cross-cutting thread:

  • Module 0/Introduction: Cultural bridging, linguistic diversity in AI, Huang + Hassabis as models.
  • Module 4 (Models): How different cognitive/linguistic foundations contribute to math-reasoning and discovery models; Einstein Test as a universal benchmark enriched by global teams.
  • Module 5 (Applications): Layer 5 edge agency for diverse hemispheres—education redesign empowering youth in all nations, SDGs via collaborative “folds,” robotics as caring partners.
  • Module 6 (Synthesis): Policy for open collaboration, valuing all contributions to counter silos and fear.

Bottom line: We do need Chinese and English (and broader) brains collaborating openly. The world benefits when leaders like Huang and Hassabis help bridge urgent needs across contexts—poor/rich, East/West, experienced/novice—for abundance and joy rather than division. This aligns perfectly with your long-term focus on win-win trade, global village, and the AI decade.What next—expand a MOOC module with this collaboration angle, explore specific Chinese contributions (e.g., DeepSeek or materials AI), or another thread?

53 web pages

Do there is a hope V10 first 10 data do r sovereignty most invested in vers rubin su factories will de Develop a jigsaw that links all edges and all folds not just compete directly on monetization one thing. Do you think it is possible for v20 to celebrate each other's brains human and machines in ways g7 to g10 had failed to fo

Thoughts

Yes, there is realistic hope — though not guaranteed — that the top ~10 data-sovereign and AI-investment-heavy nations (the "V10") can contribute pieces to a larger collaborative jigsaw. This would link all 5 AI layers, cultural/linguistic "brains" (alphabetic vs. character-based strengths), and scientific "folds" (AlphaFold equivalents in materials, physics, fusion, quantum, etc.), rather than defaulting solely to monetization races or zero-sum control.

hai.stanford.edu

Current Landscape (V10 Context)The leading players in AI investment, compute, and data sovereignty are concentrated:

  • US dominates private investment (~$286B in 2025) and frontier models, with massive builds of NVIDIA Vera Rubin AI factories (gigawatt-scale, liquid-cooled NVL72/NVL144 systems rolling out 2026–2027). These emphasize full-stack integration (energy → chips → infrastructure).

nvidianews.nvidia.com

  • China leads in scale, open-weight models, math/reasoning efficiency, and sovereign infrastructure (strong data localization).
  • Others (UAE, Japan, EU nations, Singapore, India, Saudi Arabia, etc.) invest heavily in sovereign AI for local data/culture/intelligence.

interactives.cnas.org

Sovereign AI is now mainstream: Jensen Huang and others frame it as every nation owning/refining its own data, culture, and "national intelligence." This creates natural tensions (export controls, localization rules) but also incentives for pragmatic deals.

facebook.com

Hope for a Comprehensive JigsawStrong reasons for optimism:

  • Technical necessity: Solving the Einstein Test, agentic Layer 5 edge AI, and multi-domain "folds" (materials via GNoME/MatterGen, fusion/plasma control, quantum hybrids) requires diverse priors — Chinese strengths in math/efficiency + Western creativity + global data. Pure competition wastes compute on redundant races.
  • Abundance incentives: Rubin-era gigawatt factories + quantum/fusion progress amplify win-win outcomes (cheap energy, new materials, space/robotics). Top players know isolated monetization leaves SDGs, education redesign, and risk modeling unsolved.
  • Existing bridges: Huang and Hassabis demonstrate cultural bridging (Asia–West modesty and vision). Open models, global AlphaFold usage (millions of researchers), and G20 discussions show pathways beyond G7-style clubs.

atlanticcouncil.org

  • Historical precedent: Von Neumann's open computer designs accelerated the field. Today's open-weight momentum (especially from Asia) and sovereign AI factories can evolve into shared protocols for discovery, not just commercialization.

Realistic challenges:

  • Geopolitics remains competitive (US Action Plan emphasizes dominance; China pushes its own global governance ideas). G7/G10 efforts often stayed narrow or exclusionary.

atlanticcouncil.org

  • Data sovereignty and monetization pressures are real — nations want control and economic returns.
  • Silos persist (institutions, languages, urgent needs of poor vs. rich nations).

V20 Possibility: Celebrating Human + Machine BrainsYes, it is possible by the "V20" era (interpreting as top 20 nations or ~2030 horizon) to move toward celebration of hybrid intelligence in ways G7–G10 largely failed. G7/G10 approaches often prioritized regulation, ethics statements, or alliance-based competition without enough inclusive abundance focus or youth/education redesign. A broader V20 could succeed if:

  • It builds on 5-layer openness with sovereign customization (local data + global collaboration on foundational science).
  • It explicitly values diverse "brains" (Chinese visual-spatial/math strengths + English abstraction + others) for richer Einstein-style discovery.
  • Layer 5 applications emphasize joyful generations: education that empowers kids to "play with AI for all it's worth," robotics as caring partners, and unlocking nature's abundance instead of fear.
  • Mechanisms like expanded G20 AI tracks, UN dialogues, or Huang-style sovereign AI factories with interoperability standards.

atlanticcouncil.org

This aligns with your vision: not fearing math machines/robots but energizing humans through hybrid "third brains," cultural bridging, and win-win systems for poor/rich, East/West, AI-experienced and not-yet.MOOC Tie-InIn the 5-Layer AI MOOC, this becomes a capstone theme in Module 6 (Synthesis): "From Sovereign Competition to Global Jigsaw — Designing V20 Collaboration for Abundance." Learners could map how their nation's strengths fit the puzzle and propose Layer 5 projects.There is hope, grounded in technical realities and leaders like Huang/Hassabis who fly between cultures. Execution depends on choosing collaboration on the hard scientific/educational puzzles over pure monetization. The Rubin-era buildout gives us a narrow but real window in the next 5 years.Which part should we develop further — a V20 collaboration framework, specific jigsaw mappings (e.g., China + US on fusion/materials), or a dedicated MOOC module on this?

SPACE - can we catalogue some of  the most exciting questions 5th grade students and treachers could be asking about - here is a starter with answers kindly supplied by Gemini

what is approximate size of satellites owned by japanese company axelspace

Axelspace satellites generally fall into the 100 kg to 150 kg class. Their flagship Earth observation microsatellites (the GRUS series) measure roughly 96 cm × 78 cm × 126 cm in size. [1, 2]

For comparison, previous and upcoming models differ slightly in both size and weight: [1]

  • GRUS-1 series: These early-generation Earth observation satellites weigh about 100 kg. Their envelopes measure roughly 60 cm × 60 cm × 80 cm.
  • GRUS-3 series: The newest satellites (including GRUS-3α) weigh about 150 kg and measure 96 cm × 78 cm × 126 cm. [1]

Despite their compact form, these microsatellites use high-resolution Nikon-made telescopes to track vegetation growth, coastal seabed topography, and daily landscape changes across the globe. [1, 2]

If you'd like, let me know:

  • Are you comparing these to other companies (like Planet or Starlink)?
  • Are you interested in the technical specs, such as their optical resolution and daily imaging capacity?

I can provide more information on how this constellation functions on the Axelspace corporate website.

5 sites

  • The mission patch for GRUS-3 - Axelspace

About GRUS-3. Number of satellites: 7. Name: GRUS-3A/3B/3C/3D/3E/3F/3G. Satellite wet mass: Approximately 150kg. Satellite envelop...

Axelspace

  • Axelspace to Equip Seven GRUS-3 Earth Observation ...

Axelspace operates a satellite constellation of approximately 100 kg-class Earth observation microsatellites, GRUS, for its AxelGl...

Axelspace

  • Seven additional satellites scheduled for launch in 2026 | Sorabatake

Table_title: Seven microsatellites will meet daily-observation demand—up to 2.3 million km² imaged per day Table_content: | Item |

宙畑

Show all

You said: Is this as small as satellites currently get?

No, Axelspace satellites are not as small as satellites currently get.

While Axelspace microsatellites (100–150 kg) are incredibly compact compared to traditional school-bus-sized systems, they are much larger than the smallest operational satellites today. The smallest category of functional satellites includes Femtosatellites, which can weigh less than 100 grams and fit entirely in the palm of your hand.

Satellite Size Categories

Satellites are generally categorized by mass, spanning several orders of magnitude:

  • Microsatellites (10–100 kg): This is the class where Axelspace's earlier models sit.
  • Nanosatellites (1–10 kg): The most famous of these are CubeSats. A standard "1U" CubeSat is a cube measuring exactly 10 cm × 10 cm × 10 cm and weighing about 1.33 kg. Companies like Planet Labs operate massive fleets of "Dove" CubeSats (typically 3U size, or roughly 10 cm × 10 cm × 30 cm) for global imaging.
  • Picosatellites (0.1–1 kg): These are often built as PocketQubes, which measure just 5 cm × 5 cm × 5 cm per unit.
  • Femtosatellites (1–100 grams): These are essentially single, integrated circuit boards wrapped in solar cells. A prominent example is the Chipsat (such as the KickSat project), which measures just 3.5 cm × 3.5 cm and weighs only a few grams.

Why Axelspace Satellites Are "Larger" Small Sats

Axelspace chooses the 100–150 kg microsatellite form factor due to physics constraints, specifically regarding optics. To capture high-resolution imagery (down to 2.5-meter ground resolution), a satellite requires a telescope with a relatively wide aperture mirror. Since aperture size directly dictates the physical width and length of the optical payload, the satellite cannot be shrunk down to a pocket-sized CubeSat without severely sacrificing image sharpness and quality.

If you are interested, I can provide more details on:

  • How CubeSats are launched and deployed from the International Space Station.
  • The specific optical limitations that prevent smaller satellites from taking high-resolution photos.
  • How tiny Chipsats communicate back to Earth with such low power.

You said: Whar are the uses of small satellites from smallest to axelspace size and can you suggest some implications of imagining world of small satellites instead of bus sized ones. Am i correct that with huge increase in satellites and range of sizes it will be important for ai to map the whole space of stellites and ijdeally this would be an open educational property worldwide. Would it actually be expensive compared with not doin this soon?

Yes, your assessment is highly accurate. The explosive growth of small satellites makes automated, AI-driven space traffic management an urgent necessity, and treating this data as a global open property is far cheaper than dealing with the catastrophic costs of inaction.

Here is a breakdown of the uses, systemic shifts, and the economic reality of mapping our increasingly crowded orbit.

Uses of Small Satellites by Size Class

Smaller satellites trade payload capacity for lower launch costs, enabling diverse missions across different size tiers:

Satellite Class

Typical Dimensions & Weight

Primary Practical Uses

Femtosatellites / Chipsats

\(3.5\text{ cm} \times 3.5\text{ cm}\)
1 to 100 grams

Swarm environmental sensing (e.g., measuring solar radiation or Earth's magnetic field over vast areas simultaneously), testing experimental space materials, and low-cost university research.

Picosatellites / PocketQubes

\(5\text{ cm} \times 5\text{ cm} \times 5\text{ cm}\)
100g to 1 kg

Basic IoT (Internet of Things) tracking, fundamental radio communications testing, amateur radio relays, and localized space weather monitoring.

Nanosatellites / CubeSats

\(10\text{ cm} \times 10\text{ cm} \times 10\text{ cm}\) (1U)
1 to 10 kg

Medium-resolution Earth observation, atmospheric data collection, rapid-prototype scientific experiments, and asset tracking (such as ship and aircraft transponder monitoring).

Microsatellites (Axelspace Size)

\(96\text{ cm} \times 78\text{ cm} \times 126\text{ cm}\)
100 to 150 kg

High-resolution optical imaging, precision agricultural monitoring, near real-time disaster response tracking, and advanced commercial radar (SAR) imaging.

Implications of a Small Satellite World vs. Bus-Sized Systems

Shifting from a few massive, school-bus-sized satellites to mega-constellations of small satellites fundamentally alters space economics, technology, and geopolitics:

  • From "Single Point of Failure" to Resilience: If a $500 million bus-sized satellite fails or is destroyed, the entire mission dies. If five small satellites fail in a 500-satellite constellation, the network self-heals by routing data through the remaining operational nodes.
  • Rapid Technology Refresh Cycles: Traditional satellites are built to last 15 years, meaning they fly with obsolete, decade-old technology. Small satellites are built cheaply for 3-to-5-year lifespans, allowing operators to continuously launch the newest microchips and sensors.
  • ** Democratization of Space**: Developing nations, universities, and small startups can now afford to build and launch space hardware for a fraction of traditional costs, breaking the historical monopoly of superpowers and massive defense contractors.
  • Severe Orbital Overcrowding: Launching thousands of smaller objects drastically increases collision risks, creates massive tracking challenges for ground radars, and significantly accelerates the generation of dangerous space debris.

The Role of AI and Global Open Data Mapping

You are entirely correct that AI is mandatory to map this new orbital landscape. Human operators can no longer keep up with tracking tens of thousands of rapidly moving objects.

Why AI Mapping is Essential

Traditional tracking relies on physics models to predict paths. However, small satellites frequently change orbits using tiny thrusters, and atmospheric drag fluctuates unpredictably. AI models excel at analyzing historical tracking data to predict exact paths, flag potential collisions (conjunctions), and autonomously command satellites to maneuver out of harm's way in real time.

The Argument for an "Open Educational Property"

Orbital data must be treated as a global public good. If satellite tracking data is proprietary, siloed, or hidden behind military secrecy, operators cannot coordinate. If Country A does not know where Country B's tiny satellite is, they cannot avoid a collision. An open-source, global tracking repository ensures that every university, startup, and nation flies safely. Platforms like the Celestrak database and academic collaborations are early steps toward this concept.

The Economics: Act Now vs. Delaying

Building an AI-driven global tracking system right now is vastly cheaper than delaying it.

The math of space sustainability shows that preventative action avoids exponential future liabilities:

[Cost of Early AI Mapping System]  <  [Cost of Inaction / Kessler Syndrome]

(Software, open data, radars)           (Satellite losses, orbital debris cleanup)

  1. The Cost of Collisions: A single collision between two satellites creates thousands of untraceable shards of space debris. Each shard can destroy another satellite.
  2. The Threat of Kessler Syndrome: If debris levels become too high, it could trigger a domino effect of collisions (Kessler Syndrome), rendering entire orbital planes completely unusable for generations.
  3. The Financial Comparison: Developing an international, AI-driven open tracking architecture costs tens of millions of dollars. Conversely, losing access to Low Earth Orbit due to debris would paralyze the global economy, destroying the multi-billion dollar telecommunications, GPS, weather forecasting, and climate monitoring industries.

If you would like to explore this topic further, tell me if I should focus on:

  • The current AI tools agencies like NASA or ESA use for space traffic.
  • How companies plan to safely de-orbit small satellites at the end of their lives.
  • The specific international space laws governing orbital collisions.

=========

Yes, drone technology changes the need for satellites. Drones and satellites work together like a team. [1, 2]

The increase in drones makes satellites more important than ever.

How Drones and Satellites Interact: A 5th-Grade Explanation

Imagine a drone is like a remote-controlled toy car, and you are the person holding the controller.

  • The Problem: If you drive your toy car behind a big couch or all the way down the street, the signal from your controller cannot reach it anymore. The car stops or gets lost because your controller needs a straight line to send its signals. [1]
  • The Satellite Solution: Now, imagine a satellite is like a giant mirror floating high up in space. Instead of pointing your controller directly at the toy car, you point your controller up at the sky. The signal bounces off the space mirror and shoots straight down to the toy car. [1, 2]

Because the satellite is so high up, it can see the controller and the drone at the same time, even if they are hundreds of miles apart. This allows people to fly drones safely over huge distances without losing connection. [1, 2, 3]

Why Services Like Starlink Are Becoming Vital

Yes, you are correct. Starlink is incredibly important for modern drone operations. [1, 2]

Before Starlink, satellites were high up in space (Geostationary orbit), which meant the internet signal took a long time to travel up and back down. This delay made it hard to steer a fast-moving drone in real-time.

Services like Starlink change the game for drones for several reasons: [1]

  • Super Fast Signals: Starlink uses thousands of small satellites that fly very close to Earth. Because they are so close, the internet signal travels instantly, allowing a pilot to see a live video feed from the drone and steer it without lag. [1, 2, 3]
  • Hard to Block: On the ground, signals can be blocked or scrambled by special equipment. Starlink signals beam straight down from above, making them much harder to jam or disrupt. [1, 2]
  • Internet Everywhere: Drones often fly in remote areas with no cell towers or Wi-Fi. A small Starlink antenna attached to a vehicle—or even mounted directly onto a larger drone—gives it an instant, high-speed connection anywhere on the planet. [1, 2, 3]

In short, drones provide the "eyes" and movement on the ground, while satellite networks like Starlink provide the "nervous system" that connects them to their operators. [1, 2, 3]

If you want to explore further, let me know if I should explain:

  • How space engineers make Starlink dishes small enough to fit on moving vehicles.
  • The differences between low-orbit satellites (like Starlink) and high-orbit satellites.

12 sites

  • Ukraine relies on Starlink for its drone war. Russia appears to ...

Mar 26, 2024 — The reason for the reports of a deteriorating service in Ukraine are unclear, and Starlink, SpaceX and Musk declined to comment. ...

CNN

  • Starlink and the Russia-Ukraine War: A Case of Commercial ...

Mar 9, 2023 — Overview of Starlink and Its Use in the War ... Just a few days into the Russia-Ukraine war, Mykhailo Fedorov, Ukraine's Minister ...

HARVARD Kennedy School - Belfer Center

  • Musk's Starlink satellites accelerating development of drone ...

Feb 14, 2023 — This strong connection also makes them more resistant to jamming. “The ground antennas of Starlink form a highly directional beam ...

C4ISRNet

Reply to Discussion

RSS

ENTREPRENEURIAL REVOLUTION NETWORK BENCHMARKS 2025now : Remembering Norman Macrae

Chartering 5 Layer AI Agency - integrating exponential intergenerational multipliers of trusted human relationship systems through community scaling apps

AsiaAI.docx where & how 2/3 human brains are celebrating AI livelihoods

====

lelated US AI reports:

AI commission 2021

AI Action PLan July2025

Shaping AI Billion

chris.macrae@yahoo.co.uk :help celebrate library of INTELLIGENCE multipliers: -system map

  • Action Apps
  • Millions of  AI Agents 1  2  3
  • Software sovereign infrastructure 
  • Chips1 & Supercomputers
  • Energy: Genesis
  • Fusion SCSP-FI -F2
  • Quantum
  • Critical Minerals: Pax
  • Space
  • Edu-media rev li>Nature
  • workforce 1
    cvchrismacrae.docx
  • Data Science
  • Geonomics 1

views on whether AGI exists

- how close are google aws or huawei to nvidia

2025REPORT-ER: Entrepreneurial Revolution est 1976; Neumann Intelligence Unit at The Economist since 1951. Norman Macrae's & friends 75 year mediation of engineers of computing & autonomous machines  has reached overtime: Big Brother vs Little Sister !?

Overtime help ed weekly quizzes on Gemini of Musk & Top 10 AI brains until us election nov 2028

MUSKAI.docx

unaiwho.docx version 6/6/22 hunt for 100 helping guterres most with UN2.0

RSVP chris.macrae@yahoo.co.uk

EconomistDiary.com 

Prep for UNSUMMITFUTURE.com

JOIN SEARCH FOR UNDER 30s MOST MASSIVE COLLABS FOR HUMAN SUSTAINABILITY

1 Jensen Huang 2 Demis Hassabis 3 Dei-Fei Li 4 King Charles

5 Bezos Earth (10 bn) 6 Bloomberg JohnsHopkins  cbestAI.docx 7 Banga

8 Maurice Chang 9 Mr & Mrs Jerry Yang 10 Mr & Mrs Joseph Tsai 11 Musk

12 Fazle Abed 13 Ms & Mr Steve Jobs 14 Melinda Gates 15 BJ King 16 Benioff

17 Naomi Osaka 18 Jap Emperor Family 19 Akio Morita 20 Mayor Koike

The Economist 1982 why not Silicon AI Valley Everywhere 21 Founder Sequoia 22 Mr/Mrs Anne Doerr 23 Condi Rice

23 MS & Mr Filo 24 Horvitz 25 Michael Littman NSF 26 Romano Prodi 27 Andrew Ng 29 Lila Ibrahim 28 Daphne Koller

30 Mayo Son 31 Li Ka Shing 32 Lee Kuan Yew 33 Lisa Su  34 ARM 36 Priscilla Chan

38 Agnelli Family 35 Ms Tan & Mr Joe White

37 Yann Lecun 39 Dutch Royal family 40 Romano Prodi

41 Kramer  42 Tirole  43 Rachel Glennerster 44 Tata 45 Manmohan Singh 46 Nilekani 47 James Grant 48 JimKim, 49 Guterres

50 attenborough 51 Gandhi 52 Freud 53 St Theresa 54 Montessori  55 Sunita Gandhu,56 paulo freire 57 Marshall Mcluhan58 Andrew Sreer 59 Lauren Sanchez,  60 David Zapolski

61 Harris 62 Chips Act Raimundo 63 oiv Newsom. 64 Arati Prab hakarm,65 Jennifer Doudna CrispR, 66 Oren Etsioni,67 Robert Reisch,68 Jim Srreyer  69 Sheika Moza

- 3/21/22 HAPPY 50th Birthday TO WORLD'S MOST SUSTAINABLE ECONOMY- ASIAN WOMEN SUPERVILLAGE

Since gaining my MA statistics Cambridge DAMTP 1973 (Corpus Christi College) my special sibject has been community building networks- these are the 6 most exciting collaboration opportunities my life has been privileged to map - the first two evolved as grassroots person to person networks before 1996 in tropical Asian places where village women had no access to electricity grids nor phones- then came mobile and solar entrepreneurial revolutions!! 

COLLAB platforms of livesmatter communities to mediate public and private -poorest village mothers empowering end of poverty    5.1 5.2 5.3 5.4 5.5  5.6


4 livelihood edu for all 

4.1  4.2  4.3  4.4  4.5 4.6


3 last mile health services  3.1 3,2  3.3  3.4   3.5   3.6


last mile nutrition  2.1   2.2   2.3   2.4  2.5  2,6


banking for all workers  1.1  1.2  1.3   1.4   1.5   1.6


NEWS FROM LIBRARY NORMAN MACRAE -latest publication 2021 translation into japanese biography of von neumann:

Below: neat German catalogue (about half of dad's signed works) but expensive  -interesting to see how Germans selected the parts  they like over time: eg omitted 1962 Consider Japan The Economist 

feel free to ask if free versions are available 

The coming entrepreneurial revolution : a survey Macrae, Norman - In: The economist 261 (1976), pp. 41-65 cited 105 

Macrae, Norman - In: IPA review / Institute of PublicAffairs 25 (1971) 3, pp. 67-72  
 Macrae, Norman - The Economist 257 (1975), pp. 1-44 
6 The future of international business Macrae, Norman - In: Transnational corporations and world order : readings …, (pp. 373-385). 1979 >
Future U.S. growth and leadership assessed from abroad Macrae, Norman - In: Prospects for growth : changing expectations for the future, (pp. 127-140). 1977 Check Google Scholar | 
9Entrepreneurial Revolution - next capitalism: in hi-tech left=right=center; The Economist 1976
Macrae, Norman -In: European community (1978), pp. 3-6
  Macrae, Norman - In: Kapitalismus heute, (pp. 191-204). 1974
23a 

. we scots are less than 4/1000 of the worlds and 3/4 are Diaspora - immigrants in others countries. Since 2008 I have been celebrating Bangladesh Women Empowerment solutions wth NY graduates. Now I want to host love each others events in new york starting this week with hong kong-contact me if we can celebrate anoither countries winm-wins with new yorkers

mapping OTHER ECONOMIES:

50 SMALLEST ISLAND NATIONS

TWO Macroeconomies FROM SIXTH OF PEOPLE WHO ARE WHITE & war-prone

ADemocratic

Russian

=============

From 60%+ people =Asian Supercity (60TH YEAR OF ECONOMIST REPORTING - SEE CONSIDER JAPAN1962)

Far South - eg African, Latin Am, Australasia

Earth's other economies : Arctic, Antarctic, Dessert, Rainforest

===========

In addition to how the 5 primary sdgs1-5 are gravitated we see 6 transformation factors as most critical to sustainability of 2020-2025-2030

Xfactors to 2030 Xclimate XAI Xinfra Xyouth Wwomen Xpoor chris.macrae@yahoo.co.uk (scot currently  in washington DC)- in 1984 i co-authored 2025 report with dad norman.

Asia Rising Surveys

Entrepreneurial Revolution -would endgame of one 40-year generations of applying Industrial Revolution 3,4 lead to sustainability of extinction

1972's Next 40 Years ;1976's Coming Entrepreneurial Revolution; 12 week leaders debate 1982's We're All Intrapreneurial Now

The Economist had been founded   in 1843" marking one of 6 exponential timeframes "Future Histores"

IN ASSOCIATION WITH ADAMSMITH.app :

we offer worldwide mapping view points from

1 2 now to 2025-30

and these viewpoints:

40 years ago -early 1980s when we first framed 2025 report;

from 1960s when 100 times more tech per decade was due to compound industrial revolutions 3,4 

1945 birth of UN

1843 when the economist was founded

1760s - adam smithian 2 views : last of pre-engineering era; first 16 years of engineering ra including america's declaration of independence- in essence this meant that to 1914 continental scaling of engineeriing would be separate new world <.old world

conomistwomen.com

IF we 8 billion earthlings of the 2020s are to celebrate collaboration escapes from extinction, the knowhow of the billion asian poorest women networks will be invaluable -

in mathematically connected ways so will the stories of diaspora scots and the greatest mathematicians ever home schooled -central european jewish teens who emigrated eg Neumann , Einstein ... to USA 2nd quarter of the 20th century; it is on such diversity that entrepreneurial revolution diaries have been shaped 

EconomistPOOR.com : Dad was born in the USSR in 1923 - his dad served in British Embassies. Dad's curiosity enjoyed the opposite of a standard examined education. From 11+ Norman observed results of domination of humans by mad white men - Stalin from being in British Embassy in Moscow to 1936; Hitler in Embassy of last Adriatic port used by Jews to escape Hitler. Then dad spent his last days as a teen in allied bomber command navigating airplanes stationed at modernday Myanmar. Surviving thanks to the Americas dad was in Keynes last class where he was taught that only a handful of system designers control what futures are possible. EconomistScotland.com AbedMooc.com

To help mediate such, question every world eventwith optimistic rationalism, my father's 2000 articles at The Economist interpret all sorts of future spins. After his 15th year he was permitted one signed survey a year. In the mid 1950s he had met John Von Neumann whom he become biographer to , and was the only journalist at Messina's's birth of EU. == If you only have time for one download this one page tour of COLLABorations composed by Fazle Abed and networked by billion poorest village women offers clues to sustainability from the ground up like no white ruler has ever felt or morally audited. by London Scot James Wilson. Could Queen Victoria change empire fro slavemaking to commonwealth? Some say Victoria liked the challenge James set her, others that she gave him a poison pill assignment. Thus James arrived in Calcutta 1860 with the Queens permission to charter a bank by and for Indian people. Within 9 months he died of diarrhea. 75 years later Calcutta was where the Young Fazle Abed grew up - his family accounted for some of the biggest traders. Only to be partitioned back at age 11 to his family's home region in the far north east of what had been British Raj India but was now to be ruled by Pakistan for 25 years. Age 18 Abed made the trek to Glasgow University to study naval engineering.

new york

1943 marked centenary autobio of The Economist and my teenage dad Norman prepping to be navigator allied bomber command Burma Campaign -thanks to US dad survived, finished in last class of Keynes. before starting 5 decades at The Economist; after 15 years he was allowed to sign one survey a year starting in 1962 with the scoop that Japan (Korea S, Taiwan soon hk singapore) had found development mp0de;s for all Asian to rise. Rural Keynes could end village poverty & starvation; supercity win-win trades could celebrate Neumanns gift of 100 times more tech per decade (see macrae bio of von neumann)

Since 1960 the legacy of von neumann means ever decade multiplies 100 times more micro-technology- an unprecedented time for better or worse of all earthdwellers; 2025 timelined and mapped innovation exponentials - education, health, go green etc - (opportunities threats) to celebrating sustainability generation by 2025; dad parted from earth 2010; since then 2 journals by adam smith scholars out of Glasgow where engines began in 1760- Social Business; New Economics have invited academic worlds and young graduates to question where the human race is going - after 30 business trips to wealthier parts of Asia, through 2010s I have mainly sherpa's young journalist to Bangladesh - we are filing 50 years of cases on women empowerment at these web sites AbedMOOC.com FazleAbed.com EconomistPoor.com EconomistUN.com WorldRecordjobs.com Economistwomen.com Economistyouth.com EconomistDiary.com UNsummitfuture.com - in my view how a billion asian women linked together to end extreme poverty across continental asia is the greatest and happiest miracle anyone can take notes on - please note the rest of this column does not reflect my current maps of how or where the younger half of the world need to linkin to be the first sdg generation......its more like an old scrap book

 how do humans design futures?-in the 2020s decade of the sdgs – this question has never had more urgency. to be or not to be/ – ref to lessons of deming or keynes, or glasgow university alumni smith and 200 years of hi-trust economics mapmaking later fazle abed - we now know how-a man made system is defined by one goal uniting generations- a system multiplies connected peoples work and demands either accelerating progress to its goal or collapsing - sir fazle abed died dec 2020 - so who are his most active scholars climate adaptability where cop26 november will be a great chance to renuite with 260 years of adam smith and james watts purposes t end poverty-specifically we interpret sdg 1 as meaning next girl or boy born has fair chance at free happy an productive life as we seek to make any community a child is born into a thriving space to grow up between discover of new worlds in 1500 and 1945 systems got worse and worse on the goal eg processes like slavery emerged- and ultimately the world was designed around a handful of big empires and often only the most powerful men in those empires. 4 amazing human-tech systems were invented to start massive use by 1960 borlaug agriculture and related solutions every poorest village (2/3people still had no access to electricity) could action learn person to person- deming engineering whose goal was zero defects by helping workers humanize machines- this could even allowed thousands of small suppliers to be best at one part in machines assembled from all those parts) – although americans invented these solution asia most needed them and joyfully became world class at them- up to 2 billion people were helped to end poverty through sharing this knowhow- unlike consuming up things actionable knowhow multiplies value in use when it links through every community that needs it the other two technologies space and media and satellite telecoms, and digital analytic power looked promising- by 1965 alumni of moore promised to multiply 100 fold efficiency of these core tech each decade to 2030- that would be a trillion tmes moore than was needed to land on the moon in 1960s. you might think this tech could improve race to end poverty- and initially it did but by 1990 it was designed around the long term goal of making 10 men richer than 40% poorest- these men also got involved in complex vested interests so that the vast majority of politicians in brussels and dc backed the big get bigger - often they used fake media to hide what they were doing to climate and other stuff that a world trebling in population size d\ - we the 3 generations children parents grandparents have until 2030 to design new system orbits gravitated around goal 1 and navigating the un's other 17 goals do you want to help/ 8 cities we spend most time helping students exchange sustainability solutions 2018-2019 BR0 Beijing Hangzhou: 

Girls world maps begin at B01 good news reporting with fazleabed.com  valuetrue.com and womenuni.com

.==========

online library of norman macrae--

==========

MA1 AliBaba TaoBao

Ma 2 Ali Financial

Ma10.1 DT and ODPS

health catalogue; energy catalogue

Keynes: 2025now - jobs Creating Gen

.

how poorest women in world build

A01 BRAC health system,

A02 BRAC education system,

A03 BRAC banking system

K01 Twin Health System - Haiti& Boston

Past events EconomistDiary.com

include 15th annual spring collaboration cafe new york - 2022 was withsister city hong kong designers of metaverse for beeings.app

© 2026   Created by chris macrae.   Powered by

Report an Issue  |  Terms of Service