AI inference chip startup Etched raises $700M at $21B valuation, doubling in a month

Etched, a startup building specialized chips for AI inference, has raised $700 million at a $21 billion valuation in a round led by trading firm Jane Street — roughly doubling the company's valuation in under a month after a $300 million Series C valued it at $10.3 billion in July.

What's new

According to TechCrunch's reporting on the deal, "Jane Street has installed Etched's first shipped AI cluster system, and was so impressed, it led another massive round." Jane Street had shipped in and deployed Etched's first production rack, tested it against its own trading and AI workloads, and then decided to lead the new financing itself rather than simply remain a customer.

Key details:

• Round size: $700 million, lifting Etched's valuation to $21 billion.
• Trajectory: the company was valued at $5 billion in December 2025, $10.3 billion after a $300 million Series C in July 2026, and now $21 billion — roughly doubling again in about four weeks.
• Investor list: alongside lead investor Jane Street, the round includes Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Tiger Global, Bain Capital Ventures, and Blackstone, among others.
• Product: Etched builds what it calls "frontier inference clusters" — full AI systems built around two custom components the company designed from scratch: a low-voltage prefill chip for processing prompts, and a separate memory/interconnect system tuned for the token-generation (decode) stage of inference.
• Traction: the company says it has booked more than $1 billion in customer contracts across AI labs, cloud providers, and private buyers, and has already begun shipping hardware — Jane Street's rack being the first delivered unit.

Context

Etched is one of a growing cluster of startups — alongside Groq, Cerebras, SambaNova, and others — trying to chip away at Nvidia's dominance of AI compute by building hardware purpose-built for inference rather than general-purpose training GPUs. Its pitch is architectural specialization: splitting prefill and decode into separate custom silicon rather than running both on the same general-purpose accelerator. The speed of this round — a second valuation-doubling within roughly a month, coming right after Groq closed its own $350 million Series A at a $3.5 billion valuation with NVIDIA participating — signals investors are racing to place bets across the inference-hardware field as demand for cheaper, faster model serving outstrips available capacity.

Why it matters

A trading firm leading a $700 million round in a chip startup, after first buying and stress-testing the hardware as a customer, is an unusual validation signal — it suggests Etched's inference systems perform well enough in a demanding, latency-sensitive production environment to convince a sophisticated buyer to become a lead investor rather than just a client. At $21 billion, Etched's valuation now sits closer to established inference players, underscoring how much capital is flowing into alternatives to Nvidia GPUs for serving (rather than training) large models. Whether that valuation holds will depend on Etched converting its rapid funding pace and early customer wins into broader commercial deployment against well-funded rivals and Nvidia's own inference-optimized roadmap.