Μake us preferred on Google A research team that was once approached to help lead Jeff Bezos‘s Project Prometheus has unveiled its own venture instead: an AI system built to handle an enormous volume of physics data in a single query.
The company, Accelerated Understanding Inc., says its model handled 5 trillion pieces of data in one prompt during testing, roughly 5 million times more than what today’s leading AI systems from Anthropic and Google can typically process at once.
Unlike large language models built to predict the next word in a sentence, this system is designed to predict physical phenomena across space and time. Rather than relying on the Transformer architecture that underpins most chatbots, it uses “neural operators,” a technique co-founder Anima Anandkumar helped develop years earlier. She described the approach as a shift away from a human-centered view of intelligence toward one centered on the physical world.
Several companies, including startups linked to AI researchers Yann LeCun and Fei-Fei Li, are pursuing similar “world models” that aim to understand physical space better than text-based AI. Accelerated Understanding is betting that a broadly capable neural-operator system, one that can predict phenomena invisible to the human eye, is the better path forward commercially.
The founders point to chip design as one use case: while other companies apply text-based AI reasoning to semiconductor work, they argue a deeper grasp of underlying physics could optimize materials and temperatures with far less lab trial and error. They also see potential uses in robotics, extreme-weather prediction, and analyzing geological data for energy companies, replacing narrow, custom-built math models with one general-purpose physics AI. For now, the company is targeting enterprise deals rather than consumer products.
The Bezos connection dates to late 2024, when investor and biotech entrepreneur Vik Bajaj, who later co-founded Prometheus with Bezos, discussed a potential collaboration with Anandkumar and Jenik over dinner near Los Angeles. Anandkumar, a Caltech professor who previously worked as a scientist at Amazon and served five years as an Nvidia board director, and Jenik, an AI infrastructure engineer, had already launched their own company by that point.
Bajaj later sent the pair an offer letter for what was called “Project Prometheus.” According to a copy reviewed by Reuters, it proposed Anandkumar as the company’s public face, a board member, and owner of its scientific vision, with a 35% combined equity stake for her and Jenik, plus a starting salary of $1 million a year that would double after three months. The letter also described more than $2 billion in committed financing through a Series B round, backed in part by Bezos. Prometheus declined to comment on the offer.
Anandkumar and Jenik chose to keep building independently. Bezos and Bajaj went on to raise $12 billion for Prometheus’s own Series B in June 2026, with that company focused on automating the manufacturing of complex physical systems.
Anandkumar, who declined to discuss her company’s funding, said she already has partnerships with computing providers supplying the hardware clusters behind her AI system, though she declined to name them. Nvidia did not respond to a request for comment on whether it was involved. She said Nvidia CEO Jensen Huang was the one who first encouraged her to pursue the idea, after being impressed by an earlier project she led there showing AI could match the accuracy of complex weather-forecasting computations at much greater speed. Huang later featured that work on neural operators at Nvidia’s 2021 GTC conference.
