Discovered Materials Launches Five AI Challenges in Materials Science

A young specialist from San Francisco is turning five materials science puzzles into a $1 million competition. The initiative is designed as a public testbed for the capabilities of AI models, as the startup claims to be accelerating its timeline to deliver useful materials for semiconductors.

The startup accelerates its timeline and aims for useful materials in less than a year

Advaith Sridhar claims to be more confident in Discovered Materials' ability to find industrially useful materials within a timeframe of less than a year. He notes that these timelines have slightly accelerated from the initial goal. In August, during the announcement of its funding round, he hoped to generate a useful material within a year. The company's primary objective is to develop new materials for semiconductors. It collaborates with major chip manufacturers on research projects and also funds its own scientific work.

A total prize of $1 million for five key questions

Discovered Materials is launching a $1 million challenge to solve five critical problems in materials science. The company has recruited academic advisors from top global universities to define these five open problems, in a field that lacks equivalents to the Millennium Prize Problems. The competition explicitly aims for solutions to these questions, each considered foundational for the discipline.

Targeted scientific objectives, from superconductivity to silicon interfaces

The targets cover five areas. First, a theory explaining the superconductivity of cuprates. Next, a numerical method to predict the structures of interfaces between silicon and silicon dioxide. The program is also looking for a numerical method to map grain structure related to thermal properties to design better heat dissipation materials. It includes a theory predicting why the illumination of materials can change their behavior, particularly around superconductivity. Finally, it aims for a numerical method that improves Density Functional Theory for atomic and molecular modeling.

Initial trials: several days of autonomy for AI agents, little progress

Before the announcement, cutting-edge lab models were tested on these problems. Discovered Materials allowed AI agents to use the Internet and other resources to operate autonomously for several days. The models spent several days on each problem but made little progress. Advaith Sridhar believes that tracking the attempts will show whether the models are making advancements in this area. He emphasizes that it doesn't matter who finds the solutions; the goal is for these challenges to serve as benchmarks, and solving them would deepen the understanding of materials and their behavior. It is anticipated that organizations like OpenAI or Anthropic may attempt to tackle these challenges with more computing power or newer models.

An approach designed as a public scientific benchmark

The initiative is presented as a shift in how research is conducted, as AI demonstrates its ability to discover unexpected ideas and techniques for human researchers. Advaith Sridhar explains that the first step is to establish checks and challenges, then allow explanatory avenues to emerge, noting that formulating the right questions is a challenge and that professors have participated in their development. AI models have already provided solutions to questions in mathematics and made advancements in life sciences, but they have yet to thoroughly address the complex physics of advanced materials. Discovered Materials, co-founded by Sridhar, is a San Francisco-based startup that uses AI models to identify new substances for advanced chips and was part of Y Combinator's Spring 2026 cohort. Felipe Jornada, a Stanford professor who helped develop the challenges, reminds us that observed phenomena remain unexplained and that their explanations would directly influence electronics, energy storage, and materials manufacturing.