Robots such as Figure Neo demonstrate what LeCun means when he criticises chatbots: AI that learns by manipulating real objects, not by processing text. Prometheus wants to industrialise this approach.
Yann LeCun, Chief AI Scientist at Meta, recently called Large Language Models a “dead end”: “We have language systems that pass the bar exam, but where is our domestic robot? Where is a robot as good as a cat in the physical world?”
Jeff Bezos listened. And he responded with $6.2 billion.
For the first time since 2021, the Amazon founder is returning to the operational helm of a company as co-CEO of Project Prometheus, an AI startup that raised the largest initial funding round in the industry's history — three times the previous record — before even announcing a product. The mission: to build “AI for the physical economy,” systems that learn from the real world through robotics, physical simulations, and automated experimentation. Not chatbots. Not virtual assistants. Robots that learn by doing.
The Bezos-Bajaj duo and the team of “top brains”
While Bezos brings capital and strategic vision, Vik Bajaj brings the scientific credentials that lend credibility to the ambition. The co-founder and co-CEO of Prometheus is not just any serial entrepreneur: he is a physicist-chemist with a PhD from MIT who has spent his career turning moonshots into billion-pound companies.
At Google X, Sergey Brin's “impossible factory”, Bajaj contributed to projects that became Waymo (autonomous cars, valued at £45 billion) and Wing (drone deliveries). In 2013, he co-founded Verily, Alphabet's life sciences division. As Chief Scientific Officer of GRAIL, he led the development of AI-based tests for early cancer diagnosis. Prior to Prometheus, he founded Xaira Therapeutics (AI-based drug discovery) and led Foresite Labs, an incubator specialising in AI startups for the life sciences.
His track record: transforming cutting-edge research into scalable commercial products. Exactly what is needed for a project that aims to industrialise physical AI.
The team has about 100 employees, recruited with a surgical strategy: Nal Kalchbrenner (formerly of Google Research, a pioneer in deep learning), Sherjil Ozair and William Guss (founders of General Agents, acquired by Prometheus in November 2025), and Ashish Vaswani as advisor — the lead author of the paper ‘Attention Is All You Need’ that introduced the Transformer architecture underlying ChatGPT. Operational locations: San Francisco, London, Zurich.
First strategic move: lightning-fast acquisition of General Agents
At the end of November 2025, just ten days after the New York Times revealed the existence of Prometheus, the start-up completed its first acquisition, according to Wired: General Agents, founded in 2024 by Sherjil Ozair (formerly of Google DeepMind) and William Guss (formerly of OpenAI research scientist).
The deal brings with it Ace, described as a “realtime computer pilot” — an AI agent that automates complex tasks on computers in real time. In a demo video, Ace downloads an image from Google and sends it via iMessage in less than 15 seconds. It seems trivial, but it represents a technological leap: the speed of execution that General Agents has achieved remains unattainable for its competitors.
‘What General Agents has really solved is speed — Ace runs on your computer at lightning speed,’ confirms Harsha Abegunasekara, CEO of Donely (Ace's competitor). ‘We've been working on it for six months and we haven't achieved it yet.’
But the real value isn't Ace. It's the underlying architecture.
According to General Agents' job posting, Ace is built on a VLA (Video-Language-Action) architecture — the same technology used by researchers to program industrial robots. While Ace applies it to digital tasks, VLA allows robots to:
Process visual inputs in real time
Understand natural language commands
Perform physical actions in the real world
And that's exactly what Prometheus is focused on: AI for manufacturing, industrial robotics, physical process automation.
The speed of the acquisition reveals the strategy.
Corporate filings in Delaware show that Bajaj formed the legal entity to acquire General Agents the morning after a private dinner at the two-Michelin-starred Saison restaurant in San Francisco, organised in June 2025. The deal was finalised four days later.
General Agents now operates from the headquarters of Foresite Labs (Bajaj's biotech incubator in San Francisco). Two days after the acquisition, William Guss posted on social media: ‘Looking for introductions to people working in US manufacturing. Really want to understand the space and see some factories :)’.
The message to competitors is clear: Prometheus doesn't build from scratch, it aggregates the best teams with surgical acquisitions. And it does so at Amazon speed.
$6.2 billion: redefining what “early-stage” means
The figures for Project Prometheus are unprecedented for an early-stage start-up. A “significant portion” comes directly from Bezos (who is worth £180 billion), but the other investors remain undisclosed.
To put this figure into context:
Company | Total Funding | Valuation | Stage | Notes |
|---|---|---|---|---|
Project Prometheus | $6.2B | Not disclosed | Early-stage | Before any product |
Thinking Machines Lab (Murati) | $2B | $12B | Seed | Previous record |
Safe Superintelligence (Sutskever) | $3B | $32B | Pre-product | Former OpenAI CTO |
xAI (Musk) | $22B+ | $230B | Growth | Already commercial (Grok) |
OpenAI | $57–64B | $300–500B | Late | Market leader |
Anthropic | $14–26B | $61–170B | Growth | Claude AI |
Looking ahead: Prometheus raised as much in one round as OpenAI did in its first five years of existence. The message to investors is clear: this is not a laboratory experiment, it is an industrial gamble.
Bezos had already reported his interest in physical AI by investing $400 million in Physical Intelligence (2024), a startup that develops universal “brain” software for robots. But Prometheus represents a leap in scale: from passive investor to operational co-CEO.
“AI for the physical economy”: beyond the frontier of chatbots
Project Prometheus pursues an approach that is radically different from Large Language Models. While ChatGPT learns from patterns in internet text, Prometheus builds systems that learn from direct interaction with the physical world — through robotic experiments, sensors, and simulations.
LeCun's criticism of LLMs is not isolated. Even Geoffrey Hinton (one of the “godfathers” of deep learning) has admitted: “Language models are brilliant at manipulating symbols, but they don't understand gravity, inertia, or the properties of materials.” Prometheus aims to bridge this gap.
Key technologies:
Autonomous robotic laboratories that perform thousands of experiments per day, capturing even failures (rarely published but valuable for training)
High-fidelity physical simulation (digital twins) that process millions of scenarios overnight — accelerating discovery without physical costs
VLA (Video-Language-Action) architecture acquired with General Agents: allows models to process sensory data in real time and make autonomous decisions
Closed-loop learning: AI proposes hypotheses → tests them in the real world → learns from the results → iterates
Stated target sectors:
Aerospace: spacecraft design, component optimisation for Blue Origin
Automotive: electric vehicle design, autonomous production lines
Computing: chip design, next-generation semiconductors
The synergy with Blue Origin is evident. On November 13, 2025, the New Glenn rocket successfully completed the landing of the booster — only the second company after SpaceX to do so. Bezos at Italian Tech Week 2025: “If there is work to be done on the lunar surface, we can send robots.” Prometheus could be the “brain” of those robots.
The Periodic Labs model: same game, 20x scale
The closest reference to Prometheus' approach is Periodic Labs, founded by Liam Fedus (co-creator of ChatGPT) and Ekin Dogus Cubuk (leader of DeepMind's GNoME project, which discovered 2.2 million new materials). The startup raised $300 million to build “AI scientists” with autonomous robotic laboratories.
The crucial difference: scale and focus
Periodic Labs | Project Prometheus | |
|---|---|---|
Funding | $300M | $6.2B (20×) |
Focus | Scientific discovery (materials, superconductors) | Industrial engineering and manufacturing |
Output | New molecules, materials | Commercial products, manufacturing processes |
Timeline | Fundamental research (10+ years) | Industrial applications (3–7 years) |
As one analyst notes on TechCrunch: “The internet corpus is exhausted. The real frontier is nature itself — and nature requires robots, sensors, and simulations, not just GPUs grinding through text.”
Blue Origin and Amazon: strategic synergies (but tactical independence)
The most natural connection is with Blue Origin. Fortune reports that “Project Prometheus seems integral to Bezos' vision of human expansion beyond Earth.” The VLA architecture acquired with General Agents could enable:
Autonomous manufacturing for space
Robotic assembly for lunar/orbital construction
AI optimization of the interplanetary supply chain
As for Amazon, the potential synergies are obvious but — according to analysts — deliberately limited:
Possible synergies:
Amazon as the first industrial customer (it has 1 million+ robots in 300+ logistics centers)
AWS as cloud infrastructure for simulations
Custom AI chips (Trainium/Inferentia) to reduce dependence on NVIDIA
Why independence:
Avoid conflicts with external investors (many of Amazon's competitors)
Allow partnerships with other industrial giants (GM, Boeing, etc.)
Speed of decision-making: Prometheus does not report to Amazon's board
A former Amazon executive comments off the record: "Bezos has learned from Blue Origin that some moonshots work better outside the Amazon umbrella. Too much bureaucracy, too many conflicts of interest.“
The competitive landscape: the war of billionaires and industrial giants
Elon Musk responded to the news on X with ”Haha no way 😂 Copy 🐈" — his standard insult when Bezos enters one of his markets. But the two companies are playing different games:
Aspect | xAI / Grok | Project Prometheus |
|---|---|---|
Focus | General-purpose AI, consumer chatbot | Industrial / manufacturing AI |
Current product | Grok 4, SuperGrok Heavy ($300/month) | None (pre-product) |
Target customer | Consumers, X Premium subscribers | Large enterprises (aerospace, automotive) |
Infrastructure | Colossus supercomputer (200K+ H100 GPUs) | Not disclosed (likely physical labs) |
Roadmap | Software agents, digital automation | Physical robots, embodied AI |
OpenAI focuses on software agents that “enter the digital workforce” (virtual assistants, coding, analysis). Prometheus is “physical-first” — robots, sensors, manipulation of real objects.
The real competitors: the industrial giants already present
Player | Position | Competitive Advantage |
|---|---|---|
Siemens | Established leader | Industrial Copilot ecosystem, 400K+ manufacturing customers |
Autodesk | Design software | Generative design integrated into Fusion 360 (2M+ users) |
Microsoft | Cloud platform | Azure IoT Operations, partnership with Rockwell Automation |
Boston Dynamics | Robotics hardware | Atlas/Spot robots with 15+ years of R&D, already commercial |
Siemens-Audi case study: 25x faster inference for weld inspection using Industrial AI Copilot. The result: defects detected in real time, not at the end of the line.
The key question: Will Prometheus succeed in beating incumbents with decades of B2B relationships and vertical know-how? Or will it end up like many Google X moonshots — technically brilliant but commercially irrelevant?
Implications for SMEs: distant promises, immediate alternatives
Short-term scenario (1-3 years): Direct access to Prometheus for SMEs is unlikely. The focus will be on:
Large enterprise customers (aerospace, automotive)
Strategic partnerships with Bezos' ventures (Blue Origin, potential Amazon Robotics customers)
Pilots with industrial OEMs (machinery manufacturers)
But the market is becoming more democratic:
91% of SMEs implementing AI report direct increases in revenue (McKinsey 2024 study)
Industrial robots cost 60% less than in 2017 ($27,000 → $10,856)
Physical AI market: from $3.78 billion (2024) to $67.91 billion (2034), CAGR 33%
RaaS (Robot-as-a-Service) model: eliminates capex, monthly costs comparable to a worker
Framework for SMEs: the 4 steps
Identify “quick wins”: repetitive processes with data already available (visual QC, inventory management)
Build data infrastructure: IoT sensors, shop floor digitization (cost: $10-30K)
Pilot with subscription tools: avoid massive capex, experiment with RaaS
Scale progressively: from 1 pilot line to entire plant
Medium-term scenario (3-7 years): Prometheus technology could “trickle down” to SMEs through:
Integrations with existing platforms (Siemens, Autodesk)
Cloud-based simulation tools for design/engineering
OEM partnerships: machinery with “AI Prometheus inside”
The Tulip example: founded by MIT researchers, today it serves SMEs with affordable “industrial AI” solutions. If Prometheus follows this trajectory, we could see commercial spin-offs within 5 years.
Conclusion: what happens if Bezos wins (or loses)
Project Prometheus represents the largest capital bet ever made on pre-production AI. But beyond the numbers, it raises strategic questions for the entire manufacturing sector:
If Prometheus is successful:
The cost of industrial R&D plummets: AI simulations reduce physical testing by 70-90%
New arms race: those who don't adopt physical AI risk obsolescence in 5-7 years
Concentration of power: Bezos controls the “smart factories” that produce everything — from chips to rockets
If Prometheus fails:
$6.2 billion proves that physical AI takes decades, not years
Incumbents (Siemens, Autodesk) consolidate their position with incremental approaches
Capital markets cool on pre-production mega-rounds
The real question is not technical, it is strategic: who will control the AI that designs and manufactures the physical objects of the future? An oligopoly of tech billionaires (Bezos, Musk) or European/Asian industrial giants with 100+ years of manufacturing experience?
Bezos acknowledged signs of an “industrial bubble” in AI at Italian Tech Week, adding, “When the dust settles and you see who the winners are, society still benefits from those inventions.” The name Prometheus—the Titan who stole fire from the gods to give it to humanity—evokes both progress and hubris.
With $6.2 billion, a stellar team, and a vision of merging AI with industrial robotics, Prometheus could compress decades of development into years. Or discover, like its mythological namesake, that certain fires have consequences that even a billionaire cannot control.
The game has just begun. And European SMEs would do well not to wait for the verdict before making their move.
[BONUS - 3 questions every manufacturing SME should be asking itself today]
Do we have digitized data on our production processes? Without data, AI is useless. First step: IoT sensors on critical machinery.
What is our “costly bottleneck”? Manual QC? Machine setup? Inventory? Identify the process with the most waste → faster ROI.
Can we afford to wait 5-7 years? If competitors adopt physical AI now, the gap will become unbridgeable. Better to experiment today with accessible tools than chase tomorrow.
Fabio Lauria
CEO & Founder, ELECTE
P.S. If you are interested in learning more about how AI is transforming not only space but also business on Earth, keep following this newsletter.
Welcome to the Electe Newsletter
This newsletter explores the fascinating world of artificial intelligence, explaining how it is transforming the way we live and work. We share engaging stories and surprising discoveries about AI: from the most creative applications to new emerging tools, right up to the impact these changes have on our daily lives.
You don't need to be a tech expert: through clear language and concrete examples, we transform complex concepts into compelling stories. Whether you're interested in the latest AI discoveries, the most surprising innovations, or simply want to stay up to date on technology trends, this newsletter will guide you through the wonders of artificial intelligence.
It's like having a curious and passionate guide who takes you on a weekly journey to discover the most interesting and unexpected developments in the world of AI, told in an engaging and accessible way.
Sign up now to access the complete newsletter archive. Join a community of curious minds and explorers of the future.
