YC Startup Puts $1M Bounty on Five Unsolved Physics Puzzles
By @sharedot · · 6 pages
- Startups
- AI
- Materials Science
- Venture Capital
Discovered Materials, a YC-backed chip-materials startup, is offering $1 million for AI systems to crack five open problems in materials science.
A seed-stage startup sets the grand challenges
Discovered Materials, a San Francisco startup from Y Combinator's spring 2026 batch that uses AI models to find new substances for advanced chips, announced a $1 million challenge for AI systems to solve five critical open problems in materials science, TechCrunch reports.
The company recruited an advisory group of academics at leading global universities — including Stanford professor Felipe Jornada — to spell out the problems, because materials science lacks an equivalent of the Millennium Prize Problems.
Co-founder Advaith Sridhar explained the core difficulty: '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.'
The challenge doubles as a benchmark for whether AI is actually making progress in the domain, serving as goalposts the company says anyone is welcome to chase.
The five problems frontier models failed
The prize list targets phenomena 'we observe but cannot explain,' as Jornada put it: a theory of cuprate superconductivity, where certain copper-oxide materials show zero electrical resistance at ultra-low temperatures; a numerical method to predict structures of silicon–silicon dioxide interfaces, a combination found in almost all semiconductors; a method mapping grain structure to thermal properties for better heat dissipation; a theory of why light exposure can change material behavior, including superconductivity; and an improvement to Density Functional Theory, the workhorse modeling technique.
Startup Fortune reports each problem carries a $200,000 prize. The company first turned frontier lab models loose with internet access and let them work autonomously for days — and the models made little headway on any of them.
A negative result published on purpose
What makes this surprising is the admission baked into the announcement. Plenty of AI companies claim their models are closing in on scientific discovery; Discovered Materials published its own failed experiment first, then dared outsiders — including OpenAI, Anthropic and professors — to beat it.
Startup Fortune contrasts this with Google DeepMind's GNoME, which predicted 2.2 million candidate materials at scale, noting GNoME was a prediction engine, not a system asked to close a decades-old physics question humans have failed to answer.
Sridhar told TechCrunch it is no concern whether the solver is his team, a professor or a frontier lab: 'The reason we're putting out these challenges is because we want them to serve as the goalposts.' The approach inverts scientific method — verification and challenges first, ideas later.
The stakes for AI-for-science startups
The challenge previews a new funding model for science — a seed-backed startup posting a prize and leaderboard instead of waiting through government grant cycles — and signals where venture capital may flow next, away from chatbots toward physical science.
Startup Fortune reports the company was founded by IIT Madras alumni Advaith Sridhar and Akash Ramdas and raised a $9 million seed round in August led by Lightspeed India Partners, with Y Combinator, Peak XV Partners and Paul Graham joining.
Discovered Materials partners with chip giants while funding its own research, and Sridhar told TechCrunch that timelines for producing an industrially useful material have accelerated to 'shorter than a year.'
Whether any model claims a prize will be a falsifiable data point on the state of AI for science.