Etched raises $300M at $10.3B valuation; aims for frontier-scale AI inference at lower power, higher density

Etched, an AI chip startup founded by three Harvard dropouts in 2022, raised $300 million in Series C funding at a $10.3 billion valuation, led by Sequoia Capital. The round, which Sequoia says is the highest valuation ever for a Series C from the firm, also included Andreessen Horowitz, SK Hynix, Jane Street, and Diffusion. Etched has already booked $1 billion in customer orders and employs 400 people, operating a 2-megawatt data center and a 10-megawatt production facility in Milpitas, California.

Etched built custom inference chips with two core innovations: a prefill chip running at low voltage (dramatically reducing heat and enabling higher transistor density) and a 'Cluster-Scale Memory' interconnect that allows all accelerators in a rack to share the same memory pool. This eliminates per-accelerator data duplication and improves memory efficiency for large AI workloads. The company's thesis is that inference — the compute-expensive phase after a model sees a prompt — can be split into prefill (understanding context, compute-heavy) and decode (generating output tokens, memory-heavy), each requiring different hardware optimizations.

Demand for Etched's systems reportedly outpaces supply as customers move from evaluation to production deployment. First rack shipments are scheduled for summer 2026. The company is competing in an increasingly crowded inference-focused market (Cerebras, Groq, SambaNova, Fractile) as the AI workload mix shifts: inference now dominates training as the key driver for cloud-scale AI. Architects evaluating second-source GPUs for large deployments should track Etched's production ramp as an alternative to exclusive NVIDIA reliance, particularly for latency-sensitive inference workloads.