Accelerated Understanding Inc. launches physics-centric AI model

Researchers Anima Anandkumar and Benedikt Jenik have launched Accelerated Understanding Inc., a startup featuring an AI model that departs from the standard Transformer architecture used by major industry players. Instead, the company utilizes neural operators, a framework co-pioneered by Anandkumar, to focus on a “nature-centric view” of intelligence rather than language-based processing.

During testing, the new model demonstrated the ability to process 5 trillion data points in a single prompt, a capacity approximately 5 million times larger than the context windows of leading models from Google or Anthropic. The technology is designed to predict physical phenomena in space and time, targeting enterprise sectors such as chip design, robotics, energy, and weather prediction.

The founders established the company after declining an offer to join Project Prometheus, a Jeff Bezos-backed initiative co-founded by Vik Bajaj. Project Prometheus recently secured $12 billion in Series B funding and focuses on automating the design and manufacturing of complex physical systems.