Discovered Materials, a San Francisco-based startup that has used AI models to discover new materials for manufacturing advanced chips, has announced the launch of a $1 million competition to incentivize AI models to solve five critical problems in materials science.
Despite the successes achieved by advanced AI models in solving prominent mathematical problems and their progress in life sciences, they have not yet succeeded in deciphering the complex physics associated with advanced materials. Advaith Sridhar, co-founder of the company, which participated in the Spring 2026 batch of Y Combinator, noted that clearly defining problems in materials science is extremely difficult compared to mathematics, which has definitive definitions. This prompted the company to enlist an advisory group of academics to identify five key problems.
The Five Problems Targeted in the Challenge
The five problems proposed for the cash prizes span several fundamental aspects of materials physics and modern technologies, including:
- Developing a theory to explain the superconductivity of cuprate materials, which is the ability of some materials to possess zero electrical resistance at extremely low temperatures.
- Finding a numerical method to predict the structures of interfaces between silicon and silicon dioxide, a combination essential in the manufacturing of almost all semiconductors.
- Developing a numerical method to link grain structure to thermal properties to design better materials for heat dissipation.
- Providing a theory that predicts the causality of changing material behavior when exposed to light, particularly regarding superconductivity.
- Improving density functional theory, a key technique in materials science for modeling the structures of atoms and molecules.
What Changes Practically?
Before the official announcement of these challenges, the company tested advanced laboratory models, granting them access to the internet and other tools to work independently for several days. Despite this, the models did not achieve significant progress in solving these problems. Officials explained that the goal of these challenges is not limited to who finds the solution, but rather to provide clear benchmarks for evaluating the progress of AI models in understanding the complex physics of the materials around us.
This step reflects a shift in the path of scientific research, where verification mechanisms and challenges are created first, then the field is left for models and ideas to provide scientific explanations. On the other hand, the company confirmed that its timeline for generating industrially useful materials for use in the semiconductor sector has accelerated, and it has become more confident in achieving results in less than a year.