Jeff Bezos' new company Prometheus has just completed a $12 billion financing round, with a post-investment valuation of about $41 billion. The company currently has only about 150 employees, no public product, no public customers, no revenue data, and not even a technology white paper available for external scrutiny.
With a valuation of $41 billion and 150 people, the average valuation per person is about $273 million.
This figure is not standard in the industry. In the financing records of the previous generation of AI unicorns, an average valuation exceeding $50 million per person is already considered top-tier. Prometheus has pushed this figure up by five times. It is not a normally valued company; it is a giant trust check, with the beneficiaries being its two founders and a yet-to-be-defined track.
What constitutes this check, and what is it betting on?
What's Behind the $41 Billion
To understand this valuation, we need to return to the specific conditions of two financing rounds.
According to CNBC reports, Prometheus completed its first round of financing of $6.2 billion by the end of 2025, with Bezos as the largest financial backer. Within six months, the company completed its second round of financing of $12 billion, with a post-investment valuation of about $41 billion. Axios reported on June 11 the list of investors for this round: led by Bezos, with participation from JPMorgan, Goldman Sachs, BlackRock, DST Global, and Arch Venture Partners.
This list itself indicates the logic behind the valuation's composition. JPMorgan, Goldman Sachs, and BlackRock are not typical angel or series A investors. Their presence in the shareholder list of a startup with no revenue usually indicates two things: the lead investor provided some sort of risk hedging signal; the capital size of the investment project is already large enough that traditional venture capital cannot independently bear it.
GeekWire's exclusive interview on the day of the financing 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's CEO in 2021. He is not an investor or chairman; he is a co-chief executive officer.
This choice of position has a substantive impact on the valuation. When one of the world's wealthiest individuals decides to personally manage a startup and leads the investment with his own funds, the signal he gives is more direct than any roadshow document: if this company fails, the person with the most to lose stands at the center of management. The decisions by JPMorgan and Goldman Sachs to invest are less based on a technical judgment of the physical AI track and more on an assessment of Bezos's personal credibility and risk exposure. In the stage lacking product validation, this is the closest thing to a risk control bottom line.
Co-founder Vik Bajaj provides another layer of credibility. CNBC reports that Bajaj is a co-founder of Verily, a subsidiary of Alphabet, has worked at Google X with Sergey Brin, and also serves as a professor at Stanford Medical School. His resume spans life sciences, precision engineering, and large research project management, with traceable experience in the complexity of physical systems and long-cycle R&D. Both share a common judgment: that the R&D processes for extremely complex physical systems can be reconstructed by AI. Bezos provides capital and execution will, while Bajaj provides a narrative of scientific feasibility.
The three office locations of the 150-person team also support this narrative. GeekWire and TechCrunch reported that Prometheus has offices in San Francisco, London, and Zurich. San Francisco is connected to the AI research community and venture capital, London is near global industrial engineering and financial resource centers, and Zurich is backed by academic traditions in precision manufacturing and system simulation. These three nodes correspond to talent, capital, and engineering validation, and this geographical layout itself is a signal of resource allocation before new products are launched. CNBC also reported that the team is recruiting researchers from OpenAI, Google DeepMind, and Nvidia.
An average valuation of $273 million per person is not a pricing of the current productivity of the 150 people, but a bet on the future leverage of this team. If Prometheus's path proves successful, the software produced by 150 people could replace the design hours of tens of thousands of engineers. At that point, the logic of average valuation would be completely different. But "if" is the heaviest word here.
Not Building Robots, but AI for Designing Robots
The concept used by Prometheus externally is "general artificial engineer." This term easily conjures associations with general artificial intelligence, or at least embodied intelligence, but the company has made clear delineations through multiple channels.
TechCrunch reported on June 11 that Prometheus does not create robot hardware but develops "AI for designing hardware." Bajaj provided a specific example in his interview with GeekWire: designing a jet engine, from concept to prototype to final manufacturing, usually requires a team of engineers to spend ten years or more. Prometheus aims to solve this end-to-end process as an AI problem. Application scenarios include drug molecule R&D, bridge design, chip manufacturing, etc., sharing the common feature of lengthy R&D chains, high verification costs, and trial-and-error cycles measured in years.
This positioning completely separates it from mainstream companies in the current physical AI track. Embodied intelligence addresses the execution layer of the physical world, how robots move, grasp, and operate physical objects in unstructured environments. Prometheus wants to solve the design layer of the physical world, such as how to optimize the aerodynamic layout of an engine, predict the binding energy of a drug molecule with a target protein, and how to arrange the physical layout of a chip to avoid leakage issues caused by quantum tunneling effects.
Bezos expressed a viewpoint in the same interview that has been quoted by multiple media outlets: that AI improving productivity will lead to labor shortages rather than simple unemployment. This "labor scarcity" theory is not just a sociological stance; it simultaneously paves a logical pathway for Prometheus's business model: if AI can make the design of complex physical systems ten or even a hundred times faster than it is now, but the manufacturing process still requires a huge number of engineers and technicians to execute, then the 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 $18.2 billion total financing amount across two rounds. Both CNBC and GeekWire reported that the company stated funds from this round will mainly be used to meet huge computing power demands and build specialized training data. The pixel-level simulations of fluid dynamics in jet engine combustion chambers, calculations of interactions between candidate drug molecules and proteins, and modeling of thermodynamics and electromagnetic field distributions in advanced process chips consume far more computing power than the current training requirements for large language models. If Prometheus's technical route indeed points to the path of combining physical simulation with AI, then a single round financing scale of $12 billion is not an exaggeration but rather the price of an entry ticket.
However, the specific content of the technical route has not been disclosed by the company. Whether it uses a hybrid architecture of large language models and physical simulation engines, generates physical designs directly based on diffusion models, or trains a foundational model of the physical world from scratch is completely unknown to the outside world. The team led by Fei-Fei Li previously clarified the conceptual boundaries of "world models" in a paper, distinguishing between three levels: renderer, simulator, and planner. The capabilities claimed by Prometheus conceptually point to the levels of simulator or even planner, but in the absence of any technical documents or demos made public, this direction can only remain at the conceptual level.
One Track, Two Valuation Logics
Placing Prometheus's valuation back within the physical AI track makes the comparison clear.
Data from PitchBook and Sacra show that Figure AI had a valuation of $39 billion after completing Series C financing in September 2025, with a team of about 400 to 500 people. Figure AI deals with physical entities of bipedal humanoid robots, addressing technical challenges ranging from mechanical structure, motor control, to battery management systems and human safety interactions. Its valuation is based on hardware prototypes, plant pilots, and multiple public demonstrations.
Physical Intelligence's rumored valuation is $11 billion, while Skild AI's valuation after Series C in January 2026 is between $14 billion and $15 billion. Both companies are working on general intelligence in robotics, with differences in technical architecture and ecological strategy. They are situated in the middle valuation range, lower than Figure AI and lower than Prometheus.
Prometheus's $41 billion valuation creates an inversion within the track: the company with the highest valuation has the least visible product.
The capital-led valuation rankings suggest a judgment. The robot foundational model track that Physical Intelligence and Skild AI are tapping into is relatively crowded, with OpenAI, Google DeepMind, and several Chinese companies all laying out plans, and the risk of converging technical routes is higher. Figure AI's humanoid robot path faces the triple constraints of hardware costs, yield in mass production, and safety compliance, with scaling speed strictly limited by the laws of the physical world.
Prometheus's track is design automation software that does not involve hardware manufacturing, theoretically having lower marginal costs and a higher ceiling. A design proposal for 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 capacities. If this route proves viable, the market size it can access is indeed larger than any single category of robot hardware.
But the weight of "if" becomes heavier here. Anthropic's valuation skyrocketed from $550 million to nearly $1 trillion in five years, while xAI still received exorbitant valuations despite incurring losses of $6.4 billion. These trends demonstrate that the high investment, high loss, and high valuation model in the AI field is not an isolated case. However, Anthropic and xAI deal with language models and general AI, with products that are measurable and traceable API calls. So far, no company in Prometheus's physical AI track has delivered similar levels of product validation.
The Known Unknowns
Beyond all verifiable facts, the information gap regarding Prometheus is deeper than for most companies.
Reports from TechCrunch, GeekWire, and CNBC are consistent on this point: the company has not publicly disclosed any product forms, technical architecture, demo presentations, no public commercial clients or partnership information, and no commercialization timeline. GeekWire's report quotes the company's co-founder stating that "early products are coming." However, "early" and "coming" are not defined in terms of a timeframe, nor is there clarification on whether the "product" takes the form of an API interface, a SaaS platform, or a joint R&D project.
When Bezos was asked in the interview whether he would establish an affiliated fund for acquiring manufacturing companies, GeekWire reported that he responded that Prometheus might acquire some companies and help them improve manufacturing processes. Axios used more explicit wording in its report, stating that rumors point to an affiliated merger and acquisition plan of $100 billion, but there are no SEC documents or official announcements confirming that this fund has been substantively established.
The rumored fund is worth noting not because it has been confirmed as fact, but because it forms part of the narrative around Prometheus's valuation. If a closed-loop path of "AI design plus physical manufacturing" is truly realized through acquisitions, Prometheus would no longer be a pure software company but rather a vertically integrated design and manufacturing system. However, the distance between rumors and facts remains unmeasured by any public document.
The core technical risks also remain unanswered. Whether AI can truly replace end-to-end engineering of extremely complex physical systems, such as jet engines and drug molecules, lacks consensus in both engineering and academia. The constraints of physical systems are much harsher than those of pure software systems. A large language model generating erroneous code can be rerun; an AI-generated turbine blade design could result in the crash of an aircraft if there are undetected stress concentration points. The laws of thermodynamics do not provide fault tolerance mechanisms, and material fatigue does not recognize the term "hallucination." The safety redundancy requirements of physical AI are higher from day one than any pure software AI track, yet Prometheus has not publicly demonstrated its capability boundaries in handling such constraints.
This company has 150 employees, $18.2 billion in financing, two founders with impeccable resumes, three strategically chosen office locations, and an ambition capable of redefining the conceptual framework of industrial R&D. What it lacks is everything necessary for outside observers to independently assess its prospects.
This makes the essence of Prometheus's current stage exceptionally clear: it is carrying a technology vision that spans over ten years on a balance sheet in an extremely early stage. Bezos's personal wealth and credibility provide a rare protective layer for this vision, but that protective layer is not a product, and the amount of financing is not engineering validation. Whether this is the next-generation operating system for industrial R&D or the largest one-way trust bet in the physical AI track will only be revealed after the phrase "early products are coming" is fulfilled.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
