Project Prometheus quietly moved from rumor to reality as one of the most heavily funded AI startups on record, with about $6.2 billion in backing and a mandate to tackle challenging physical problems in manufacturing, aerospace, and advanced hardware.
The venture marked Jeff Bezos's return as a hands-on chief executive, this time in partnership with scientist and entrepreneur Vik Bajaj.
Behind the large funding round sat an even quieter move, the acquisition of agentic computing startup General Agents, whose software could operate computers and tools on behalf of a user.
The deal, tied to a private San Francisco dinner, signaled that Prometheus wanted AI systems capable of controlling real machines, not just generating text or images.
What is Project Prometheus trying to build
Project Prometheus emerged as a stealth AI company focused on what some investors call physical AI, systems that learn from experimentation and help design and build complex products such as computers, cars, robots, and spacecraft.
Reports described plans to apply AI to engineering workflows and industrial processes that still rely heavily on human judgment.
With roughly $6.2 billion already committed, Prometheus instantly joined the top tier of AI funding, surpassing many other early-stage labs in available capital.
The startup aimed to use that war chest for compute, specialized hardware, and teams that blend robotics, simulation, and agentic software tailored to factories and advanced manufacturing lines.
Did you know?
Project Prometheus is already one of the best funded early stage AI startups in the world, with more capital than many public tech companies raised in their IPO years
Why did Prometheus buy agentic startup General Agents?
General Agents built agentic computing tools that take a high-level instruction and operate a computer to complete multi-step tasks across different applications.
Its flagship product, Ace, launched in 2025 as a computer pilot that could navigate software on behalf of a user, a capability that fit directly into Prometheus' focus on AI that acts in the world.
Corporate records showed that a new entity tied to Vik Bajaj was formed shortly after an off-the-record AI dinner in San Francisco to acquire General Agents, then merged with the startup a few days later.
The acquisition brought not only technology but also a small team experienced in building agents that control existing tools rather than requiring custom hardware from scratch.
Who are the key figures behind this AI gamble
Jeff Bezos, now serving as co-chief executive of Project Prometheus, brought decades of experience in scaling Amazon and backing space venture Blue Origin, along with fresh investments in robotics-focused AI such as Physical Intelligence.
His involvement signaled that the new company sat at the intersection of online infrastructure, logistics, and heavy industry.
Vik Bajaj, an MIT-trained scientist and serial founder, previously helped create Alphabet's life sciences arm, Verily, led the cancer-detection company Grail, and built Foresite Labs as a biotech and data incubator.
His background in precision health, machine learning, and venture building positioned him to run day-to-day technical and scientific strategy at Prometheus.
ALSO READ | Apple Faces New EU Regulations on Ads and Maps Services
How could agentic AI reshape factories and hardware?
Agentic AI systems, such as those developed at General Agents, are designed to take language-based goals and coordinate software, files, and tools to achieve them with minimal supervision.
Applied to factories, similar agents could prepare designs, schedule production runs, run simulations, and adjust machine parameters more frequently than human operators can.
Prometheus's interest in computers, automobiles, and aerospace hinted at plans to embed AI into both design and production phases, from chip layout to robotics on assembly lines.
If successful, such systems might shorten development cycles, reduce waste, and allow rapid customization, while also demanding new safety and oversight mechanisms on the shop floor.
What risks and questions does this quiet deal raise
The combination of a large war chest, agentic software, and a focus on factories raised concerns about the pace of automation and its effect on skilled manufacturing workers.
Unions, regulators, and policymakers may soon ask whether oversight tools and liability rules can keep up with AI that can directly control industrial systems.
Another open question centers on the concentration of power in AI infrastructure, since few startups can match multi-billion-dollar funding at such an early stage.
Competitors and partners will watch closely to see whether Prometheus shares standards, data interfaces, and safety practices, or keeps its stack tightly controlled alongside Bezos' other ventures, such as Blue Origin.
Prometheus's early hiring spree, which drew talent from leading labs including OpenAI, DeepMind, Tesla, and Meta, suggested that the company sought a mix of frontier-model expertise and robotics experience.
The arrival of General Agents' founders, Sherjil Ozair and William Guss, added specialist knowledge in building agents that operate existing computer interfaces, a skill that could be translated to industrial control environments.
Two days after the General Agents acquisition, cofounder William Guss publicly asked for introductions to people working in United States manufacturing, a signal that the team wanted direct exposure to factories, plants, and actual workflows rather than staying in purely digital settings.
That outreach highlighted a broader theme across Prometheus efforts, an emphasis on grounding AI systems in real-world constraints and processes.
Project Prometheus now stands at a rare intersection of capital, talent, and industrial ambition, with the General Agents deal providing an early glimpse into how the startup might bridge software agents and physical machines.
As manufacturing, aerospace, and hardware companies look for partners, the way Prometheus deploys its war chest may shape not only factory floors but also global norms for how far AI should be allowed to act on the real world.
Comments (0)
Please sign in to leave a comment