Bezos Expands Project Prometheus With $10B Fundraise

Jeff Bezos is scaling his AI lab, Project Prometheus, toward a near-term funding close of $10 billion at a $38 billion valuation, with institutional backers including BlackRock and JPMorgan. The lab launched in November 2025 with $6.2 billion in initial funding and now runs with over 120 employees recruited from top AI labs. Leadership includes co-CEO Vikram Bajaj, with co-founders Sherjil Ozair and William Guss in technical roles.

Prometheus is positioning itself on "physical AI", systems that model and reason about real-world physics, materials, and industrial processes rather than producing natural-language outputs. Practitioners should note these core focuses:

• •building models that integrate physics priors and interaction data from real-world systems
• •targeting engineering and manufacturing workflows for computers, automobiles, spacecraft, robotics, and logistics
• •leveraging proprietary data on material behavior, tolerances, and process flows, which differ from large public text corpora

The firm is assembling cross-disciplinary teams from ML research, robotics, simulation, and domain engineering. Expect investments in physics-informed architectures, differentiable simulation, large multimodal training pipelines, and edge-to-cloud deployment stacks tailored to industrial control and CAD/data formats.

This raise, if completed, makes Prometheus one of the most highly valued early-stage AI companies and signals investor confidence in verticalized, physics-aware AI. There are three sectoral implications for practitioners. First, vertical specialization is resurging: models trained on proprietary industrial datasets can deliver practical ROI faster than generalist LLMs in manufacturing and AEC (architecture, engineering, construction). Second, the proposed separate holding company targeting up to $100 billion of acquisitions creates an unusual closed-loop strategy, where acquired firms supply domain data and contracts that accelerate model performance and commercial adoption. Third, talent flow from OpenAI, xAI, Meta, and DeepMind into a Bezos-led operation underscores competition for researchers who can bridge ML and real-world systems engineering.

Companies building industrial AI should reassess data strategy, simulation fidelity, and IP partnerships. Access to high-quality industrial datasets and the ability to run realistic differentiable simulations will be gating factors. Existing players in AEC and manufacturing could face consolidation pressure or become strategic data partners for models that need decades of process knowledge.

The key questions are whether the funding closes at scale, how Prometheus structures its data pipelines and IP governance when acquiring firms, and what specific technical stack choices it makes for simulation, digital twins, and control. Also monitor commercial pilots and regulatory scrutiny around data concentration and safety when models interact with physical systems. Rapid capital plus proprietary data could accelerate adoption, but integrating ML models into safety-critical industrial workflows will demand rigorous validation and standards.

This is a high-conviction bet on vertical, physics-aware AI with deep integration into industrial value chains. For ML engineers, product leaders, and infrastructure teams, the return to domain-specialized model development and the emphasis on proprietary industrial data changes partnership, acquisition, and data governance priorities for the next 24 months.