> ## Content Index
> Fetch the complete content index at: https://www.metatalks.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# The founders who turned down Bezos-backed Prometheus have built an AI without the transformer
- URL: https://www.metatalks.ai/founders-who-turned-down-bezos-backed-prometheus-unveil-a-transformer-free-ai-model/
- Published: 2026-08-27T15:47:00.000Z
- Updated: 2026-08-27T15:46:59.000Z
- Author: Al
- Tags: News, Frontier Models, Science AI, #newswire

**Accelerated Understanding Inc launched on Tuesday with an AI model trained to predict how physical events unfold in space and time rather than which word comes next.**

In place of the Transformer architecture Google invented and ChatGPT is named after, it routes physics through neural operators — a technology its co-founder Anima Anandkumar helped create.

Anandkumar and her husband, Benedikt Jenik, were already building the company in late 2024 when Vik Bajaj, who later co-founded [Project Prometheus with Jeff Bezos](https://www.breakingviews.com/columns/considered-view/jeff-bezos-plays-it-slow-fast-moving-ai-2026-05-08/?ref=metatalks.ai), approached them about joining forces. The offer letter that followed would have made Anandkumar the public face of Prometheus, with a 35% stake for the couple against more than $2 billion the letter called committed.

They kept building on their own. Prometheus closed a $12 billion Series B in June and declined to comment, [Reuters](https://www.reuters.com/business/ai-founders-who-walked-away-bezos-backed-prometheus-model-universe-2026-08-25/?ref=metatalks.ai) reported.

In trials it ran itself, the company says, the model took in 5 trillion data points from a single prompt — roughly 5 million times what the flagship models from Anthropic and Google handle at once.

Anandkumar, who teaches computing and mathematical sciences at Caltech, calls putting physics at the center of intelligence a nature-centric view. The company is selling to businesses first and points at chip design: where others apply AI that reasons through semiconductor problems in text, it holds that a built-in command of physics is what settles a chip’s materials and heat with fewer rounds in the lab.

Startups run by Yann LeCun and Fei-Fei Li are after the same ground.

The method traces back to Nvidia, which hired Anandkumar in 2018 to lead a team working out how its graphics chips could serve frontier AI. Jensen Huang presented her neural-operator work at the company’s GTC conference in 2021, after an early project forecast weather as accurately as the calculations forecasters have long relied on.

Huang was the one who pushed her to take up the idea, she said. When she remarked that the AI could eat physics theorists’ lunch, he replied: “I want it to eat all their lunches.”

Anandkumar would not discuss how her own company is funded, saying only that compute providers supplied the hardware clusters behind the model. Asked whether it was a backer, Nvidia did not reply.