SAN FRANCISCO, Aug 25 (Reuters) – A research duo, once pitched to lead the Jeff Bezos-backed Project Prometheus, on Tuesday unveiled what they’ve built on their own: an AI model that can take in and produce information so vast that a lone computer might not handle it.
Their company, Accelerated Understanding Inc, says it has developed a different type of AI model that in tests handled 5 trillion pieces of data in a single prompt. That is some 5 million times the size of what Anthropic and Google’s flagship models can typically consume — like reading Tolstoy’s “War and Peace” not once but 5 million times in one sitting.
The difference is that this new system is not built for understanding language. It is built for physics, said co-founders Anima Anandkumar and Benedikt Jenik, in an exclusive interview ahead of Tuesday’s corporate launch.
While systems like ChatGPT were created with the world’s text data, letting them predict the right next word in a sentence when talking to users, this alternative AI has learned to predict physical phenomena in space and time. It dispensed with the Google-invented Transformer architecture — the “T” in ChatGPT — and instead processes physics data based on technology that Anandkumar helped pioneer years ago, called neural operators.
“The language-centric view of intelligence is humans at the center. Putting physics at the center is a nature-centric view,” said Anandkumar, a Caltech professor of computing and mathematical sciences.
Various companies, including startups overseen by AI leaders Yann LeCun and Fei-Fei Li, are pursuing so-called world models that understand spatial reality better than AI trained on text. Accelerated Understanding’s bet is that a more generally capable form of neural operator that predicts phenomena even the eye can’t seeis the best path — and can boost business.
One example is in chip design. While various companies are using AI that programs computers and reasons through text for semiconductor applications, Accelerated Understanding believes that a more intrinsic grasp of physics is key to optimizing materials and temperatures for a chip’s performance, with less trial and error in a lab.
The company also thinks the same AI can power robotics, predict extreme weather, or sift through geological data for energy companies. Instead of brittle and bespoke math models for each, one AI can handle any physics query for the business world, Anandkumar said.
The company is focused on enterprise deals rather than a consumer offering to start, the co-founders said.
BEZOS-BACKED OFFER
A similar idea has interested some in the business world, including Bezos, Amazon’s billionaire founder.
Over dinner at an upscale restaurant in greater Los Angeles in late 2024, an investor and biotech entrepreneur named Vik Bajaj — who would go on to co-found Prometheus with Bezos — discussed a collaboration with Anandkumar and Jenik, according to meeting records seen by Reuters.
Anandkumar had previously worked as a scientist at Amazon, where she had attended intimate AI summits Bezos had hosted, and she had been a director at Nvidia for five years. Jenik, her husband, is an AI infrastructure engineer. The two had already started their company.
Bajaj later discussed an offer letter with them entitled “Project Prometheus.” In a copy seen by Reuters, the proposal said Anandkumar could be the public face of the company, a board member, and owner of its scientific vision. She and Jenik, to be a board observer, could get a 35% stake in the company, plus a combined $1 million annual salary that would double to $2 million after three months of work.
The offer letter outlined more than $2 billion in capital for what the letter called “committed rounds” of financing through Series B, from investors including Bezos.
Prometheus declined to comment.
Anandkumar and Jenik ultimately kept building solo, while Bezos and Bajaj went on to raise a $12 billion Series B for Prometheus in June 2026. Prometheus is targeting AI that can automate the manufacturing of complex physical systems.
ORIGINS AT NVIDIA
Though Anandkumar declined to discuss funding, she already has partnerships with computing providers who furnished hardware clusters to develop and run Accelerated Understanding’s AI, she said. She declined to name the partners.
Nvidia did not reply to a Reuters question asking if it was backing the endeavor. Nvidia CEO Jensen Huang was the one who encouraged Anandkumar to pursue the idea in the first place, she said.
Hired by the chip designer in 2018, Anandkumar led a team of scientists who pushed forward on how Nvidia’s graphics processing units, or GPUs, could be used for frontier AI.
An early project showed how AI could speed up weather prediction with as much accuracy as the complex computations forecasters used. The results amazed Huang, who presented Anandkumar’s work on neural operators at Nvidia’s annual GTC Conference in 2021.
“He just got so excited,” Anandkumar said.When she mentioned how AI could eat physics theorists’ lunch, Huang replied: “I want it to eat all their lunches.”
(Reporting by Jeffrey Dastin in San Francisco; Editing by Kenneth Li and Rosalba O’Brien)
