The Next AI Frontier: New Startup Processes 5 Trillion Data Points for Physics Simulation

Artificial intelligence researchers Anima Anandkumar and Benedikt Jenik have launched a new AI company focused on modelling physical phenomena rather than primarily processing human language. Their company, Accelerated Understanding Inc, has developed a system that its founders claim can process as many as 5 trillion pieces of data in a single prompt, targeting applications across industries that rely heavily on physics and scientific computing.

Unlike language-focused systems such as ChatGPT, the new model is designed to predict physical phenomena across space and time. The founders have moved away from the Transformer architecture widely used in modern generative AI and are instead using neural operators, a technology that Anandkumar has helped advance. She described the approach as shifting the focus of AI from a language-centred view of intelligence towards one based on the physical world.

Accelerated Understanding is positioning the technology for enterprise applications rather than an initial consumer product. The company believes its approach could be particularly useful in areas such as semiconductor design, where understanding materials, temperatures and other physical conditions could reduce the amount of laboratory trial and error required. Other potential applications include robotics, extreme-weather prediction and analysis of geological information for energy companies.

Also Read: National Teachers Awards 2026: 21 Higher Education, Polytechnic Faculty To Be Honoured

The company’s decision to develop the technology independently came after Anandkumar and Jenik were approached about joining Project Prometheus, the AI venture backed by Amazon founder Jeff Bezos. According to meeting records reviewed by Reuters, an offer would have given Anandkumar a prominent scientific and leadership role, while she and Jenik were offered a combined 35% stake and annual compensation starting at $1 million. The proposal also outlined more than $2 billion in planned financing through Series B.

The researchers ultimately chose to continue building Accelerated Understanding on their own, while Bezos and investor Vik Bajaj proceeded with Prometheus. Prometheus later raised $12 billion in a Series B round in June 2026 and is focused on AI designed to automate the manufacturing of complex physical systems. Anandkumar has not disclosed funding details for her company but said it has partnerships with computing providers that supplied hardware clusters for developing and running its AI.

Anandkumar’s work on physics-focused AI also has roots in her time at Nvidia, where she joined in 2018 and led research into using graphics processing units for advanced AI applications. One early project demonstrated how AI could accelerate weather prediction while matching the accuracy of complex conventional calculations. Nvidia CEO Jensen Huang encouraged Anandkumar to pursue the broader idea, helping strengthen the case for neural operators as a potential foundation for AI systems designed to model the physical world.

Also Read: US Military C-17 Lands In Moscow, Kremlin Says It Had No Information About Flight