The public AI conversation is dominated by language: assistants, copilots, search, document analysis and generative content. But one of the clearest signals about the next stage of AI is appearing somewhere else: scientific and industrial computation.
On June 17, 2026, SandboxAQ announced a definitive agreement with the U.S. Department of Commerce’s CHIPS Research and Development Office for a $500 million award focused on critical semiconductor materials and chemistries, including PFAS-free process chemicals, catalysts, rare-earth-free magnets and battery systems.
The important point is not that every organisation needs the same scale of programme. The signal is that quantitative AI is increasingly being treated as strategic infrastructure.
The real problem is expensive evidence
Semiconductors, batteries, catalysts and advanced materials are governed by chemistry, physics, process constraints and measurable properties. A language model can help researchers read papers or operate tools, but the decisive work requires models that evaluate the quantitative world.
SandboxAQ describes a simulation-first approach in which high-fidelity physics generates training data, quantitative models screen large candidate spaces, and strong candidates move toward validation. Its AQVolt26 release, for example, describes 322,656 high-fidelity DFT calculations for lithium-halide electrolyte modelling.
The opportunity is not “replace scientists with AI.” It is to build a better decision system around expensive quantitative work.
LargeQuant operates around the same general decision loop at the infrastructure level: objective → model → execution → evidence → uncertainty → decision → next experiment.