Discovered Materials, a months-old San Francisco startup that uses AI models to discover new materials to build advanced chips, has announced a $1 million challenge to use AI models to solve five important problems in materials science.
Frontier AI models have solved an impressive set of mathematical problems and made advances in the life sciences, but they have yet to truly unravel the complex physics of cutting-edge materials.
“In materials science, it’s more difficult to clearly specify the problem because it’s not like mathematics, where you have very specific definitions and yes-or-no decisions,” said Advaith Sridhar, co-founder of Discovered Materials, part of Y Combinator’s Spring 2026 batch. “We came to this discussion (and wondered): What are the big challenges in this space? How do we know that AI is making progress?”
Although there is no equivalent to a Millennium Prize problem in materials science, Discovered Materials employed an advisory group of academics from the world’s leading universities to identify five open questions. The company is offering prizes worth a total of $1 million for solutions to the problem.
Copper oxide superconductivity, the theory that explains the ability of certain materials to reduce their electrical resistance to zero at extremely low temperatures. A numerical method to predict the structure of the interface between silicon and silicon dioxide, a combination of materials found in almost all semiconductors. Numerical methods that map particle structure to thermal properties enable the design of better heat-dissipating materials. A theory that predicts why when a material is exposed to light, its behavior changes, especially with respect to superconductivity. Numerical methods to improve density functional theory, an important technique used in materials science to model the structure of atoms and molecules.
“Across materials science, there are phenomena that we observe but cannot explain,” Felipe Jornada, a professor at Stanford University who helped solve the challenge, said in a statement. “Explaining these phenomena will improve our understanding of physics, and the answers will directly shape how we build electronics, store energy, and make materials of the future.”
Before announcing the challenge, Discovered Materials unlocked the Frontier Lab model and provided the AI agent with internet access and other resources, allowing it to work autonomously for several days. The models spent several days working on each problem, but did not make much progress. Still, it’s easy to imagine OpenAI and Anthropic trying to solve them in one fell swoop with massively increased compute and new models.
“It doesn’t really matter to us whether we find it, someone else finds it, a professor finds it, a frontier lab finds it,” Sridhar said. “The reason we publish these challenges is because we hope they will serve as goalposts. By going to solve these challenges, we will deepen our understanding of materials and how things around us behave.”
Tracking efforts to solve the problem will tell “if the model is moving up in this area,” he added.
This challenge also reflects the changing nature of scientific investigation, as AI has proven capable of discovering insights and techniques that human researchers had not anticipated. The problem is becoming more important because it seems easier than ever to find a solution.
“This is a new way of thinking about science, where we come up with the test and the problem first, and then we ask someone else to come up with an idea that explains the problem,” Sridhar said. “In this field, it’s not easy to know the right questions to ask, it’s very difficult. So in this field, the professors really helped us shape what are the right questions to ask.”
The primary focus of Discovered Materials is the development of new materials for use in semiconductors. The startup partners with semiconductor giants on research projects while also funding its own scientific efforts. When the company announced its seed round in August, Sridhar told TechCrunch that he hoped the company would produce useful material within a year.
“Those timelines have sped up a little bit,” he says today. “I’m now confident that we can find these industrially useful materials on a timeline of less than a year.”
If you make a purchase through links in our articles, we may earn a small commission. This does not affect editorial independence.
