$12B Series B Values Prometheus' Physical AI Agents at $41B

Jeff Bezos has officially launched Prometheus from stealth, announcing a $12 billion Series B funding round that values the AI startup at $41 billion. The company is developing what Bezos calls an “artificial general engineer,” a suite of models built to automate the design and manufacturing of complex physical products.

Bezos led the $12 billion round alongside a syndicate of major financial institutions, including JPMorgan Chase, Goldman Sachs, BlackRock, DST Global, and Arch Venture Partners. The Series B brings Prometheus’ total funding to over $18 billion, following an unannounced $6.2 billion initial investment in late 2025.

Bezos will serve as co-CEO alongside Vik Bajaj, a former Google executive and co-founder of Verily. The operational role marks Bezos’ first CEO position since leaving Amazon in 2021. The company currently operates with roughly 150 employees distributed across offices in San Francisco, London, and Zurich.

The Artificial General Engineer

While incumbent AI labs have focused on reasoning in text, image, and software domains, Prometheus is exclusively targeting “Physical AI.” The platform is designed to apply artificial intelligence directly to the laws of physics and materials science.

Prometheus aims to compress industrial engineering cycles by a factor of 10x or more. During the launch, Bezos characterized the current manufacturing pipeline as a “Dream-Build Loop” constrained by manual testing. He highlighted that modifying a jet engine to yield a 10% thrust increase currently requires a 10-year development program. Prometheus intends to shrink these timelines to months by orchestrating multi-agent systems that simulate physical constraints and generate viable schematics autonomously.

The system’s initial scope covers end-to-end development for aerospace components, robotics, medical devices, semiconductors, and pharmaceutical compounds.

Training on the Physical World

Unlike large language models trained on internet text, an artificial general engineer requires high-fidelity, proprietary manufacturing telemetry and physics data. Prometheus has not disclosed its exact network architecture, but the models are reportedly trained heavily on deterministic physics laws alongside data licensed from undisclosed industrial partners.

To secure the necessary training environment, Bezos is reportedly establishing a $100 billion investment fund aimed at acquiring traditional industrial companies. These acquisitions would serve a dual purpose: providing proprietary manufacturing data for model training and acting as direct testing grounds for Prometheus deployments. This vertical integration strategy bypasses the data-licensing bottlenecks currently facing digital AI developers.

Operational Scope and Liability

Prometheus maintains no formal corporate ties to Amazon or Blue Origin. However, Blue Origin is positioned to be a primary customer for the aerospace manufacturing tools, particularly following a highly publicized rocket explosion during testing on May 28, 2026.

The introduction of generative models to physical engineering introduces strict regulatory and safety constraints. Structural engineering communities have raised immediate skepticism regarding liability. While AI agents can accelerate fluid dynamics calculations or material stress simulations, evaluating AI output for structural integrity carries lethal consequences if hallucinations occur in the design of a bridge or a jet turbine. Human engineers remain legally responsible for signing off on all generated blueprints.

Bezos dismissed concerns regarding AI-driven unemployment in the engineering sector, predicting instead that the models will create a severe labor shortage. By drastically lowering the cost and time required to design physical goods, the anticipated productivity gains are projected to increase total manufacturing volume, driving up the demand for human operators and reviewers to oversee the automated pipelines.

If you build industrial software or manage manufacturing pipelines, the shift toward physical AI indicates that deterministic CAD and simulation tools will increasingly be orchestrated by agentic models. Teams should begin evaluating how their proprietary telemetry and testing data can be structured to support in-house model fine-tuning or secure integration with external physical reasoning engines.