Etched, a startup building specialized chips for AI inference, raised $700 million in new funding led by Jane Street, with Kleiner Perkins, Sequoia, Andreessen Horowitz, Tiger Global, Bain Capital Ventures, and other investors also participating, pushing the company's valuation to $21 billion. That's roughly double what Etched was worth just a month earlier, when it raised a $300 million Series C at a $10.3 billion valuation on July 23, 2026, an unusually fast re-rating even by the standards of the current AI funding environment. Alongside the raise, Etched announced it has completed its first customer hardware delivery, shipping a rack of its inference chips to Jane Street, the same firm that led this round, suggesting the lead investor's involvement is at least partly informed by direct, hands-on testing of the product rather than valuation momentum alone. Etched's core bet is architectural specialization: rather than building general-purpose GPUs like Nvidia, the company designs chips purpose-built specifically for transformer-model inference, wagering that giving up general-purpose flexibility in exchange for efficiency on the specific math transformer models actually run will produce meaningfully better throughput and cost-per-token than general accelerators can match. Whether that bet pays off commercially depends heavily on how much of the market keeps consolidating around transformer-style architectures versus moving toward more heterogeneous model designs, since a chip that's brilliant at one architecture becomes a liability if the industry shifts. For teams building AI-heavy products, Etched's rapid valuation growth and now-proven ability to ship real hardware to a paying customer is one more data point that inference cost and throughput, not just model capability, have become one of the most heavily contested and well-funded layers of AI infrastructure, alongside efforts from Cerebras, Groq, and others chasing the same fundamental problem: making inference cheaper and faster than general-purpose GPUs can deliver it.