Jeff Bezos said Prometheus AI operates at a finer level of detail than conventional large language models by using unique training data sets [1, 2].

This shift in approach represents a move away from the text-based patterns used by most current AI. By focusing on the physical world rather than just language, Prometheus aims to create tools capable of high-precision engineering and design.

During a CNBC interview on June 11, Bezos said the startup utilizes a training data set that is completely different from those powering standard LLMs [1, 2]. He said the company must create and access data sets that are very hard to obtain to build AI that understands the physical world [1, 3].

Bezos said Project Prometheus is an artificial general engineer building next-generation design tools for physical objects [3]. This focus on physicality requires a level of detail that text-based models cannot achieve. He said, "Prometheus AI is operating at a different level of detail than LLMs" [1].

The venture has seen significant financial backing since its launch in November 2025 [4]. The company's valuation has reached $38 billion [4], following initial funding of $6.2 billion [4]. Reports also indicate a target for an additional funding round close to $10 billion [5].

Bezos serves as co-CEO of the company alongside Vik Bajaj [1, 2]. While many AI firms focus on generative text or images, Prometheus is positioning itself as a specialized tool for engineering. Bezos said, "We have to create our data sets and access data sets that are very hard to access" [1].

“Prometheus AI is operating at a different level of detail than LLMs.”

The strategy marks a pivot from the 'big data' approach of scraping the public internet toward a 'deep data' model. By prioritizing hard-to-access, specialized data sets, Prometheus is attempting to solve the 'grounding' problem in AI—the gap between digital linguistic patterns and the laws of physics. If successful, this could move AI from a creative assistant to a functional tool for industrial manufacturing and hardware engineering.