Y Combinator Startup Discovered Materials Puts $1 Million Behind Five Problems AI Models Cannot Solve

Discovered Materials, a months-old San Francisco startup that uses AI models to find new substances for advanced chips, has announced a million-dollar challenge: a prize pool worth $1 million in total for solutions to five unsolved problems in materials science. The company, which graduated from Y Combinator's spring 2026 batch, recruited an advisory group of academics at leading global universities to spell out the problems — a field that, unlike mathematics, has no Millennium Prize Problems to rally around.

Co-founder Advaith Sridhar said the difficulty starts with problem definition itself. "It's more difficult to specify problems clearly in materials science because it's not like math, where you have a very specific definition and can get to a yes or no," he said. "We came to this discussion wondering, what are the grand challenges in this field? How do we know AI is making progress?"

The five challenges span theory and numerical methods: a theory explaining cuprate superconductivity, the phenomenon where certain materials conduct electricity with zero resistance at ultra-low temperatures; a numerical method to predict the structures of interfaces between silicon and silicon dioxide, the material combination found in almost every semiconductor; a numerical method mapping grain structure to thermal properties to design better heat-dissipation materials; a theory predicting why shining light on materials can change their behavior, particularly around superconductivity; and a numerical method to improve Density Functional Theory, the workhorse technique for modeling atoms and molecules.

"Across materials science, there are phenomena we observe but cannot explain," said Felipe Jornada, a Stanford professor who helped develop the challenges. "Explaining these phenomena would improve our understanding of physics, and the answers would directly shape how we build electronics, store energy, and manufacture the materials of the future."

Before announcing the challenge, the startup stress-tested it on frontier AI models. The company turned the models loose with internet access and other resources, letting them work autonomously for days on each problem. The models made little headway — though Sridhar acknowledged it is easy to imagine OpenAI or Anthropic one-shotting the problems later with far more compute or a newer model. He said the point is not who solves them. "The reason we're putting out these challenges is because we want them to serve as the goalposts," he said.

The exercise also reflects a changing model of scientific inquiry, where AI can surface insights human researchers did not anticipate. "This is a new way of thinking about science, where we come up with the verification and the challenges first, and then we let somebody come up with the ideas that explain the problem," Sridhar said. Knowing which questions to ask is itself hard — which is where the professors helped, he added.

Discovered Materials' core business is developing new materials for semiconductors, partnering with chip giants on research while funding its own science. When it announced its seed round in August, Sridhar said he hoped to produce a useful material within a year. "Those timelines have accelerated a little bit," he said this week. "I'm more confident now that we will be able to find these industrially useful materials in a timeline shorter than a year."

The challenge doubles as a public benchmark for whether AI models are genuinely climbing in physics-heavy domains, where progress is harder to measure than in math or code. If a frontier lab, a professor or an independent researcher collects the money, the field gains both an answer and a data point on where machine scientific reasoning actually stands.