Wafer, a San Francisco startup building specialized silicon for generative AI inference, has closed a $40 million Series A round co-led by Marathon and Chemistry. Wing Venture Capital, AMD Ventures, Outset Capital, Fifty Years and Y Combinator all joined the round, disclosed on September 3.
Silicon Plus A Serving Stack, Not Just A Chip
Co-founded by Zonathan Zhang and Pranav Sharma, Wafer is pitching a hardware-and-software bundle rather than a bare accelerator. The company builds custom inference engines, high-throughput execution kernels and a deep software stack that spans heterogeneous chip clusters, along with continuous benchmarking tools that test enterprise workloads directly against Wafer's proprietary silicon. The goal Wafer keeps hammering is per-token cost: cut it, without giving up throughput, for the AI-native companies scaling to production.
Where The $40M Goes
Wafer said it will spend the Series A on high-performance compute and wafer-scale manufacturing procurement, hire aggressively across its engineering organization, and stand up a go-to-market team to onboard customers into what it describes as an asset-heavy hardware-and-software model.
AMD's Strategic Signal
The most interesting name on the cap table is AMD Ventures. The corporate arm of the world's second-largest AI accelerator vendor putting a check into an inference-only startup - one that talks about running heterogeneous chip clusters - reads as a hedge against a market where cost-per-token, not raw FLOPS, will decide which silicon actually gets deployed. Related coverage: Lasso's CPU-only AI guardrail, iPronics' NVIDIA-led $125M photonic networking round, and Enflame's oversubscribed Shanghai AI-chip IPO.
Reporting based on coverage from FinSMEs, Wafer, Marathon and Chemistry.