Radical AI's self-driving lab hits 1,200 alloys in six months

In this Latent Space episode, Brandon Anderson interviews Radical AI founder Joseph Krause, a materials scientist who built his company around the insight that no single AI model can "one-shot" materials discovery. Unlike biological molecules that can be represented as token strings, materials science involves complex macro variables — supply chains, microstructures, and manufacturing processes — that make the field uniquely resistant to purely model-driven approaches. Krause points to the LK99 episode of 2023 as an illustration: even when basic ingredients were known, undisclosed manufacturing details defeated reproducibility.

He estimates the system can scale further to produce and characterize a hundred new alloys per day.

Radical's answer is the "self-driving lab" (SDL): a closed-loop system pairing an AI scientist — which combines scientific knowledge, computational techniques, and human intuition — with automated robotics that synthesize and characterize materials. Research campaigns run in parallel rather than serially, enabling Radical to produce and characterize 1,200 alloys in six months, a pace Krause describes as nearly 10x faster than the DARPA/GE MACH program's target of 500 new alloys in a year. He estimates the system can scale further to produce and characterize a hundred new alloys per day. In one campaign, the AI scientist proposed and tested 300 new materials; ten were found to have novel state-of-the-art properties and are already being developed for commercial applications. Notably, the AI scientist has expanded into elemental and alloy families that had no prior published research, which Krause says also helps address supply chain bottlenecks for vital industries. Krause, who spent time in Washington D.C. before founding Radical, also flags China's centralized manufacturing model as a competitive threat given its ability to rapidly scale new materials from lab to production.

Key facts

1. 01Radical AI's self-driving lab produced and characterized 1,200 alloys in six months, nearly 10x the pace of the DARPA/GE MACH program (which targeted 500 new alloys in a year).
2. 02Krause estimates the system can scale to produce and characterize a hundred new alloys tested per day.
3. 03In one research campaign, the AI scientist proposed and tested 300 new materials; 10 showed novel state-of-the-art properties now being developed for commercial applications.
4. 04The SDL operates as a closed-loop system: an AI scientist generates hypotheses, and automated robotics synthesize and characterize materials in parallel rather than serially.
5. 05Krause argues that experimental data — not the AI model — is the true competitive moat in materials science.
6. 06The AI scientist has expanded into elemental and alloy families with no prior published research, with potential supply chain implications.
7. 07Krause cites the 2023 LK99 episode as evidence that undisclosed manufacturing processes can defeat reproducibility even when ingredients are known.

Topics