Jeff Bezos's new company, Prometheus, has just completed a $12 billion funding round, valuing the company at approximately $41 billion post-money. The company currently has only about 150 employees, no publicly disclosed products, no publicly disclosed customers, no revenue data, and not even a technical white paper available for external scrutiny.

The 150 people represent a $41 billion valuation, with an average valuation of approximately $273 million per person.

This figure is not industry norm. In the funding records of the previous generation of AI unicorns, an average valuation of over $50 million per person was already considered top-tier. Prometheus has pushed that number fivefold. It's not a company with a normal valuation; it's a huge check of trust, redeemed by two founders and a still undefined sector.

What is this check made of, and what is it bet on?

What is the composition of 41 billion?

To understand this valuation, we need to go back to the specific conditions of the two rounds of financing.

According to a CNBC report, Prometheus completed its first round of financing of $6.2 billion by the end of 2025, with Bezos being the largest investor. Within six months, the company completed a second round of financing of $12 billion, valuing the company at approximately $41 billion post-money. Axios disclosed the list of investors in this round in a report on June 11: Bezos personally led the investment, with JPMorgan Chase, Goldman Sachs, BlackRock, DST Global, and Arch Venture Partners participating.

This list itself illustrates the logic behind the valuation. JPMorgan Chase, Goldman Sachs, and BlackRock are not typical angel or Series A investors. Their presence on the shareholder list of a startup with no revenue usually means two things: the lead investor is providing some kind of risk hedging signal; or the invested project's capital size has reached a scale that traditional venture capital firms would find difficult to handle independently.

In an exclusive interview on the day of the funding round, GeekWire revealed a key fact: this is the first time Jeff Bezos has personally served as co-CEO of a company since stepping down as Amazon CEO in 2021. He is not an investor, not a chairman, but a co-CEO.

This career choice has a substantial impact on valuation. When one of the world's richest individuals decides to personally manage a startup and lead an investment round with his own funds, the signal is more direct than any roadshow document: if the company fails, the person who will suffer the most is at the heart of management. JPMorgan Chase and Goldman Sachs' decision to follow suit was less based on a technological assessment of the physical AI sector and more on an assessment of Bezos's personal credit and risk exposure. In the stage lacking product validation, this is the closest thing to the bottom line of risk control.

Co-founder Vik Bajaj provides the other end of the credit. CNBC reports that Bajaj is the co-founder of Verily, a subsidiary of Alphabet, and previously worked with Sergey Brin at Google X, while also serving as a professor at Stanford Medical School. His resume spans life sciences, precision engineering, and the management of large-scale research projects, with traceable experience in the complexity of physical systems and long-cycle R&D. The two share a common belief: the development process of extremely complex physical systems can be restructured by AI. Bezos provides capital and the will to execute, while Bajaj provides a feasible narrative for scientific engineering.

The three office locations of the 150-person team also support this narrative. Both GeekWire and TechCrunch reported that Prometheus has offices in San Francisco, London, and Zurich. San Francisco connects with the AI research community and venture capital, London is close to global industrial engineering and financial resources, and Zurich is backed by an academic tradition of precision manufacturing and systems simulation. These three nodes correspond to talent, funding, and engineering validation, respectively; this geographical layout itself signals resource allocation even before a new product is launched. CNBC also reported that the team is recruiting researchers from OpenAI, Google DeepMind, and Nvidia.

The valuation of $273 million per employee isn't a price tag on the current productivity of these 150 people, but rather a bet on the team's future leverage. If Prometheus's path succeeds, the software produced by these 150 people could replace the design hours of tens of thousands of engineers. At that point, the valuation per employee would be entirely different. But "if" is the most crucial word here.

AI that designs robots, not robots.

Prometheus uses the concept of "general artificial engineer" to describe itself. This term easily evokes images of general artificial intelligence, or at least embodied intelligence, but the company has clearly defined these boundaries across multiple channels.

A TechCrunch report on June 11 stated that Prometheus doesn't manufacture robotic hardware, but rather develops "AI for designing hardware." In an interview with GeekWire, Bajaj gave a concrete example: the process of designing, prototyping, and finally manufacturing a jet engine typically takes an engineering team ten years or more. Prometheus attempts to solve this end-to-end process as an AI problem. Its applications cover drug molecule development, bridge design, and chip manufacturing, all characterized by extremely long physical system development chains, extremely high verification costs, and trial-and-error cycles measured in years.

This positioning completely distinguishes it from the mainstream companies in the current physical AI field. Embodied intelligence solves the execution layer of the physical world: how robots move, grasp, and manipulate physical objects in unstructured environments. Prometheus aims to solve the design layer of the physical world: how to optimize the aerodynamic layout of engines, how to predict the binding energy of drug molecules to target proteins, and how to lay out the physical layout of chips to avoid leakage problems caused by quantum tunneling effects.

In the same interview, Bezos raised a point that has been cited by multiple media outlets: AI's increased productivity will lead to a shortage of human labor, rather than simply unemployment. This "labor scarcity" argument is not merely a sociological stance; it also paves a logical path for Prometheus's business model: if AI can make the design of complex physical systems 10 to 100 times faster than it is now, but the manufacturing process still requires a large number of engineers and skilled workers, then companies that master design automation tools will become the bottleneck resource for the entire industry chain.

The demand for computing power is another clue to understanding the total $18.2 billion raised in two rounds of funding. Both CNBC and GeekWire reported that the company stated this round of funding will primarily be used to meet its massive computing power needs and build specialized training data. Pixel-by-pixel simulations of physical processes such as jet engine combustion chamber fluid dynamics simulations, calculations of interactions between candidate drug molecules and proteins, and modeling of thermodynamic and electromagnetic field distributions in advanced process chips consume computing power far exceeding the training requirements of current large language models. If Prometheus's technological roadmap indeed points to a combination of physical simulation and AI, then the $12 billion single-round funding is not an exaggeration, but rather the price of an entry ticket.

However, the company has not disclosed the specifics of its technical approach. Whether it uses a hybrid architecture of a large language model and a physics simulation engine, directly generates physical designs based on a diffusion model, or trains a basic model of the physical world from scratch, remains completely unknown to the outside world. Fei-Fei Li's team previously clarified the conceptual boundaries of the "world model" in a paper, distinguishing between three levels: renderer, simulator, and planner. Prometheus's claimed capabilities conceptually point to the simulator or even planner level, but without any publicly available technical documentation or demos, this direction remains purely conceptual.

One sector, two valuation logics

Placing Prometheus's valuation back into the realm of physical AI makes the comparison clearer.

According to data from PitchBook and Sacra, Figure AI's valuation was $39 billion after its Series C funding round in September 2025, with a team of approximately 400 to 500 people. Figure AI develops physical prototypes of bipedal humanoid robots, facing technological challenges ranging from mechanical structures and motor control to battery management systems and safe human-robot interaction. Its valuation is based on hardware prototypes, factory pilots, and multiple rounds of public demonstrations.

Physical Intelligence is rumored to be valued at $11 billion, while Skild AI is valued at $14-15 billion after its Series C funding round in January 2026. Both companies work on general intelligence for robots, but differ in their technological architecture and ecosystem strategies. They are in the middle of their valuation range, lower than Figure AI and also lower than Prometheus.

Prometheus's $41 billion valuation creates an inversion within the industry: the company with the highest valuation has the least visible product.

The valuation rankings given by investors suggest a judgment. The competitive landscape in the basic robot modeling field, which Physical Intelligence and Skild AI are entering, is already relatively crowded, with OpenAI, Google DeepMind, and many Chinese companies all making moves, increasing the risk of convergence in technological approaches. Figure AI's humanoid robot path faces triple constraints: hardware costs, mass production yield, and safety compliance, with its scaling speed strictly limited by the laws of the physical world.

Prometheus operates in the field of design automation software, which does not involve hardware manufacturing. Theoretically, this results in lower marginal costs and a higher ceiling for growth. The design of a jet engine can be licensed to all engine manufacturers globally, and an AI platform for drug molecule design can serve all pharmaceutical companies, eliminating physical bottlenecks related to hardware supply chains and factory capacity. If this path proves successful, the market it can reach is indeed larger than that of any single category of robotic hardware.

But the "if" factor takes on greater weight here. Anthropic's valuation soared from $550 million to nearly a trillion dollars in five years, and xAI maintained a sky-high valuation despite losses of $6.4 billion. These two curves illustrate that the high-investment, high-loss, high-valuation model in the AI field is not an isolated case. However, Anthropic and xAI focus on language models and general AI, already possessing measurable products and traceable API calls. Prometheus's physics AI track currently lacks a company with similar level of product validation.

Known Unknown

Beyond all the verifiable facts, Prometheus has a deeper information gap than most companies.

TechCrunch, GeekWire, and CNBC reports are consistent on this point: the company has not disclosed any product form, technical architecture, or demo, nor has it released information on commercial customers or partners, or provided a commercialization timeline. GeekWire's report quotes a company co-founder as saying "early rollouts are coming." However, "early" and "coming" lack a defined timeframe, and the form of the "product"—whether it's an API, a SaaS platform, or a joint development project—is also unclear.

In an interview, Bezos was asked whether he would establish an affiliated fund to acquire manufacturing companies. GeekWire reported that Bezos responded that Prometheus might acquire some companies and help them improve their manufacturing processes. Axios used more specific wording in its report, stating that there were rumors pointing to a $100 billion affiliated M&A plan, but no SEC filings or official announcements confirmed that the fund had actually been established.

This rumored fund is noteworthy not because it has been confirmed, but because it forms part of Prometheus' valuation narrative. If a closed-loop path of "AI design plus physical manufacturing" is indeed achieved through mergers and acquisitions, Prometheus will no longer be a pure software company, but a vertically integrated design and manufacturing system. However, the distance between rumor and fact is currently immeasurable by any publicly available documents.

The core technological risks also remain unanswered. Whether AI can truly replace end-to-end engineering of extremely complex physical systems like jet engines and drug molecules is a matter of debate within both engineering and academia. The constraints of physical systems are far more severe than those of purely software systems. A large language model generating incorrect code can be re-run; however, an AI-generated turbine blade design with undetected stress concentration points could result in an aircraft crash. The laws of thermodynamics offer no fault tolerance, and material fatigue does not tolerate the concept of "illusion." The safety redundancy requirements of physical AI have been higher from day one than any purely software AI field, and Prometheus has yet to publicly demonstrate the limits of its capabilities in handling these constraints.

This company boasts 150 employees, $18.2 billion in funding, two impeccably qualified founders, three carefully chosen offices, and a conceptual framework ambitious enough to redefine industrial R&D. What it lacks is everything that allows outsiders to independently assess its prospects.

This makes the essence of Prometheus at its current stage exceptionally clear: it's using an extremely early-stage balance sheet to support a technological vision spanning over a decade. Bezos's personal wealth and credit provide a rare shield for this vision, but this shield isn't the product, nor is the funding amount engineering validation. Whether this is the operating system for next-generation industrial R&D or the biggest one-way trust bet in the physical AI field, the answer isn't in the valuation figures, but will only emerge after the promise of "early-stage product launch" is fulfilled.

Chance Studios, a collectible card game platform, has raised $3.2 million in funding.

Popular Articles