Fly Labs
Ask FlyBot

Discovered Materials (YC P26) – AI agents to discover new materials

Hey HN, we're Advaith and Akash from Discovered Materials ( https://discoveredmaterials.com/ ). We build AI agents that discover new materials for the semiconductor industry.GPUs today have a heat problem. Nvidia & AMD are almost doubling the TDP (Thermal Design Power) in every chip they release - the H100 (released 2022) has a TDP of 700W, Blackwell (2024) gives out 1.2 kW and Rubin (2026) gives out at 2.3 kW of heat. This trend is expected to continue, and getting rid of this heat is one of the major reasons datacenters consume so much power and water today - they need it to keep chips cool during operation.The amount of heat produced by a chip and its ability to dissipate it are both influenced by the materials used to make it. For example, we could reduce the energy per bit required to move data between logic and memory by 10-50x by 3D packaging chips (placing HBM memory stacks directly on top of logic chips, instead of placing them beside logic on a 2D circuit board). However, we're unable to do this today because the dielectric material used in HBM (such as SiO2) is a very poor thermal conductor, trapping heat between logic and memory and causing drastic temperature rise during operation. Similarly, there's many other materials in the GPU that are being re-evaluated today - 2 more examples are thermal interface materials and substrates. However, getting a new material into a fab takes years and hundreds of millions of dollars of research - the infamous "lab-to-fab valley of death".At Discovered Materials, we're optimistic that AI agents can reduce the timeline and cost required to introduce new materials into semiconductor chips. We're seeing glimpses of this already - we tested 7 models from Anthropic, OpenAI and Kimi, and found that they're all able to computationally discover new materials that are dynamically stable and possess promising properties. This was surprising to us - it would generally t

Hacker NewsToolAISource
0Sign in to voteCopy link

FL score

72

out of 100

Verdict

VALIDATE

high confidence

Competition

No competitor data yet

Trend

No signal yet

Use AI agents to computationally discover semiconductor materials that conduct heat better, compressing years of R&D into months and reducing the cost barrier for fabs to adopt them.

The pain

GPU heat output is doubling every 2-3 years (700W to 2.3kW). Current thermal materials like SiO2 trap heat in 3D chip stacks. Getting new materials into production takes 5-10 years and $100M+ in research. Datacenters waste enormous power and water cooling chips that could run cooler with better materials.

The gap

Materials science discovery is bottlenecked by slow, expensive lab iteration. LLMs can now generate thermally stable material candidates computationally. The gap is between 'AI can suggest candidates' and 'fabs will actually test and adopt them.' No one is currently bridging computational discovery to fab validation at scale.

Build angle

Start by using AI to generate 50-100 material candidates for specific problems (thermal dielectrics, interface materials, substrates). Partner with 1-2 fabs or materials labs to validate top candidates experimentally. Prove that AI-discovered materials outperform baseline options in real testing. Charge fabs per validated material or license the discovery platform. Expand to other semiconductor pain points (power delivery, signal integrity).

Strengths

  • Massive TAM: Every major fab and GPU maker faces the heat problem. Nvidia, TSMC, Samsung all need solutions.
  • Clear ROI: Reducing thermal resistance by even 10% saves datacenters millions in cooling costs annually.
  • Technical validation exists: They tested 7 LLM models and confirmed they can generate dynamically stable candidates, this is not theoretical.
  • Founders have credibility: YC P26 backing signals serious vetting.
  • Defensible moat: First-mover advantage in AI-to-fab pipeline; relationships with fabs are sticky.

Risks

  • Lab-to-fab valley is real: Computational discovery is only step 1. Experimental validation, safety testing, and fab integration take years. They may discover materials fabs won't actually use.
  • Fab relationships are hard: TSMC, Samsung, Intel have entrenched material suppliers and long qualification cycles. Selling to them requires proof of concept at their scale, not just lab results.
  • Validation cost is hidden: They haven't disclosed who's testing candidates or how much that costs. If validation is expensive, the unit economics break.
  • Competitive threat: Major materials companies (DuPont, Cabot, Entegris) and fabs' internal R&D teams could adopt the same LLM approach. No IP moat on the discovery method itself.
  • Unclear revenue model: 'Charge per validated material' assumes fabs will pay upfront for unproven candidates. More likely they want risk-sharing or royalties on adoption.
  • Narrow initial TAM: Thermal materials for HBM is a specific problem. If that gets solved by one material, revenue drops. Need multiple material classes to sustain.

Questions about this idea?

FlyBot reads the scoring and gives you a second opinion on “Discovered Materials (YC P26) – AI agents to discover new materials”.

Open FlyBot