Etched, a San Jose-based AI inference chip startup, announced it has raised $700 million at a $21 billion valuation, led by quantitative trading firm Jane Street, in a round that doubled the company's valuation in less than a month. The previous Series C closed in July at $10.3 billion pre-money, making this jump from $10.3B to $21B in roughly 30 days one of the fastest revaluations in AI hardware. Jane Street is not just an investor—it is Etched's first paying customer, having received the company's first shipped inference rack last month and now running production workloads through it. Other investors include Kleiner Perkins, Sequoia, Andreessen Horowitz, Tiger Global, Bain Capital Ventures, and others.
Etched builds specialized systems for AI inference (the runtime step where models generate outputs) around two custom-designed components: a low-voltage prefill chip and a cluster-scale memory interconnect for the decode stage. The company claims its architecture delivers higher token throughput at lower latency than general-purpose GPUs. It emerged from stealth in June with over $1 billion in signed customer contracts across AI labs, cloud providers, and infrastructure companies. The startup has achieved first-pass silicon success in under three years from seed and completed tape-out on TSMC 4nm process.
For architects evaluating inference clusters, this repricing is significant. Jane Street—operating infrastructure where microsecond latencies translate directly to trading P&L—installed and validated the hardware on real production workloads, then led the round. That sequence (test → purchase → lead invest) is distinct from paper benchmarks and signals that transformer-specific inference silicon can compete on economics against commodity GPUs. The risk: Etched must navigate manufacturing scale, delivery commitments, and long-term customer retention. Semiconductor history is littered with brilliant chips that never scaled. The $1B in contracts and Jane Street's operational validation are concrete data points, but full proof comes from multiple customers publicly reporting sustained deployments.