Startup Discovered Materials offers $1 million reward for AI solutions to five materials science problems
The startup Discovered Materials (YC, Spring 2026) has set up a $1 million fund for solving five open problems in materials science. According to the company, tests of frontier AI models on these tasks have so far yielded only minimal progress.
Discovered Materials, a San Francisco startup that is part of Y Combinator's Spring 2026 batch and develops new materials for advanced chips using AI models, announced a challenge with a total prize pool of $1 million. The reward is for solving five selected open problems in materials science, which the company says have no equivalent in this field to the Millennium Prize Problems in mathematics. The tasks were defined by an advisory group of academics from leading universities, including Stanford professor Felipe Jornada.
The five problems include: a theory explaining cuprate superconductivity, a numerical method for predicting the interface structure of silicon and silicon dioxide (Si/SiO2) used in nearly all semiconductors, a method mapping grain structure to the thermal properties of materials, a theory describing why illuminating a material can change its properties (particularly superconductivity), and a numerical method to improve Density Functional Theory, a key technique for modeling the structure of atoms and molecules.
According to Discovered Materials, before the challenge was announced, frontier AI models were deployed on these problems with access to the internet and other resources, working autonomously for several days on each task, but according to co-founder Advaith Sridhar, they achieved only minimal progress. The company says the goal of the challenge is to create measurable targets by which progress of AI models in this domain can be tracked, regardless of who ultimately solves a given problem first.
The company focuses on developing new materials for semiconductors, collaborates with chip manufacturers on research projects, and raised a seed round in August 2026. According to Sridhar, the estimated time to find an industrially usable material has shortened since August, and the company is now confident it will achieve this in under a year.
Why it matters
The challenge offers a concrete, publicly trackable benchmark of how far today's frontier AI models are from being able to independently solve complex physics problems outside mathematics and biology. The results (or lack thereof) will serve as an indicator for companies and researchers considering deploying AI agents in materials and semiconductor research.
Relevant practical impact
What this means
For a business
For companies in semiconductor and materials research, the relevant finding is that frontier AI models, given multi-day autonomous work and internet access, achieved only minimal progress on these five tasks – AI agents therefore cannot yet be reliably deployed to independently solve open physics problems in this field.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.