Why would two AI researchers walk away from the chance to help lead a Jeff Bezos-backed AI company? Anima Anandkumar and Benedikt Jenik selected to seek their own technical vision and their startup, Accelerated Understanding Inc., has now introduced an AI model designed to understand physics instead of language.
The company says its system processed 5-trillion pieces of data within a single prompt during testing. That scale far exceeds the typical context windows related with leading large language models.
Moreover, the comparison only tells part of the story. Increased Understanding isn’t trying to build another ChatGPT competitor. Its model targets physical systems across space and time.
Why the Founders Walked Away From Bezos-Backed Prometheus
Anandkumar and Jenik had already commenced building their company after they entered discussions with biotech entrepreneur Vik Bajaj, who later co-founded Project Prometheus with Amazon founder Jeff Bezos.
As per to documents reviewed by Reuters, a proposed Project Prometheus offer would have positioned Anandkumar as a public face, board member, and leader of the company’s scientific vision. Jenik would have served as a wide observer.
The proposal reportedly included a integrated 35% ownership stake for the pair and outlined more than $2 billion in committed financing through a Series B round, including of investment from Bezos.
Anandkumar and Jenik ultimately declined and persisted developing Accelerated Understanding independently. Prometheus later raised a reported $12 billion Series B in June 2026 because it pursued AI systems capable to automating the producing of complex physical products.
Accelerated Understanding Puts Physics Before Language
The technological difference between the 2 founders’ vision and mainstream generative AI begins with architecture. Large language models commonly depend upon Transformers to predict sequences of text. Accelerated Understanding rather uses neural operators, technology Anandkumar supported pioneer for modeling physical phenomena.
Instead of anticipating the next word, the system learns relationships within physical systems and predicts how those systems evolve across space and time.
Anandkumar explains this as a shift from a human-centered view of intelligence toward one centered on nature and physics.
The approach also places Accelerated Understanding within the developing field of world models, where AI researchers are working to construct systems capable of understanding physical environments instead of primarily processing language.
Physics AI Could Target Chips, Weather, Robotics, and Energy
Accelerated Understanding sees semiconductor engineering as one in its potential enterprise applications. A physics-aware AI model could support engineers model materials, thermal conditions, and other variables affecting chip performance earlier than depending on extensive physical experimentation.
The same underlying technology could assist robotics, geological analysis, energy exploration, and extreme weather prediction.
The company’s wider ambition is to update highly specialized mathematical models with a more general AI system capable of managing multiple categories of physics issues.
For now, Accelerated Understanding plans to focus on enterprise deployments instead of consumer applications.
The Technology Has Roots at NVIDIA
Anandkumar’s work on neural operators predates the startup. After joining NVIDIA in 2018, she led researchers investigating how GPUs could accelerate frontier AI workloads. Her team’s work showed that neural operators could dramatically accelerate weather prediction maintaining competitive accuracy.
NVIDIA CEO Jensen Huang later emphasized the research during the company’s 2021 GTC conference and encouraged Anandkumar to persist exploring the technology.
Accelerated Understanding is now attempting to turn those ideas into a general-purpose foundation for physics-focused AI.
