$1 Million AI Challenge Tackles Five Materials Science Mysteries

AI has cracked math puzzles and mapped proteins, but the physics of advanced materials remains stubbornly out of reach. San Francisco startup Discovered Materials, a Y Combinator spring 2026 cohort, is betting that a $1 million prize will force the next generation of models to tackle the field’s toughest questions.

The company hand‑picked five open problems with the help of top university professors, then let state‑of‑the‑art AI agents roam the internet for days. The models made little headway, a stark reminder that materials science is not a tidy, yes‑or‑no discipline like mathematics.

Yet the startup sees the lack of progress not as a failure but as a benchmark: if a model can solve one of the five, it will prove that AI is finally grasping the complex physics that govern how atoms arrange themselves. What makes this challenge different is that Discovered Materials is using AI to set the questions, not just the answers.

By publishing the problems first, the company turns the usual scientific workflow on its head.

Researchers and companies can now measure whether their models are truly climbing the materials science ladder, and the results will inform how future AI systems are trained.

The stakes are high.

Solving any of the five would deepen our understanding of physics and directly influence how we build electronics, store energy, and manufacture tomorrow’s materials.

For the AI community, it is a litmus test of whether the next wave of models can move beyond pattern recognition into genuine scientific insight.

Keep an eye on the competition: the first model to crack a problem could rewrite the rules of both AI and materials science.