matforge is building autonomous AI scientists that hunt for the next generation of semiconductor materials. Power consumption and heat release from AI chips is roughly doubling each year, and the industry needs better interconnect, dielectric, and substrate materials to keep Moore's Law alive. Today that search takes ten or more years of grinding lab work. matforge wants to compress the cycle into months by orchestrating a swarm of AI agents across the full discovery loop.
The agents propose candidate materials, design synthesis routes, run computational screens, and coordinate physical lab experiments to validate the most promising leads. The platform is designed to ingest results, update its internal world model, and iterate without waiting for a human PI to interpret each batch. By closing the loop between simulation and wet lab, matforge aims to deliver 10x better alternatives for nanoscale interconnects and other critical layers used by foundries such as Intel and TSMC.
Founded in 2026 by Akash Ramdas, a Stanford PhD whose own discoveries for nanoscale interconnects have been adopted into Intel and TSMC roadmaps, and Advaith Sridhar, a former founding applied scientist at Persona AI who shipped long-horizon autonomous agents at scale, matforge is backed by Y Combinator in the Spring 2026 P26 batch. Headquartered in San Francisco, the company targets foundries, IDMs, and datacenter chip designers who feel the materials ceiling first.