Wafer

Wafer is an artificial intelligence infrastructure company that builds autonomous performance engineering agents to optimize GPU kernels for AI inference workloads. In practice, that means their core product takes PyTorch code and automatically generates custom kernels that optimize GPU usage, enabling teams without specialized GPU engineers to achieve significant speed and efficiency gains. The product is aimed at teams running open source LLMs in production who want the throughput of a hand tuned inference stack without hiring a dedicated kernel team, and the pitch behind the "keeps getting better" tagline is that the agents keep profiling and rewriting kernels as models, hardware, and traffic patterns shift. The company was founded in 2025 by Emilio Andere and Steven Arellano and based in San Francisco, Wafer emerged from the Y Combinator Summer 2025 batch . Arellano previously worked at Two Sigma, Google, and Sei Labs, while Andere came from Argonne, the UChicago SAND Lab, and Elicit, with both holding degrees from the University of Chicago. Earlier this year they closed a $4M seed round led by Fifty Years with participation from Liquid2 and Y Combinator. This launch marks a jump in ambition. The founders set out to raise $18M and instead closed a $40M Series A co-led by Marathon Management Partners and Chemistry, which lands as demand for cost efficient inference across both Nvidia and AMD accelerators is spiking and hyperscalers are racing to squeeze more tokens per watt out of existing fleets. For founders and operators shopping for inference, Wafer is worth watching as an alternative to the standard managed serving stacks, one where the optimization layer is itself an agent that improves over time.