Etched, a startup building specialized chips for AI inference, announced Tuesday that it has raised $700 million at a $21 billion valuation, led by quantitative trading firm Jane Street. The jump is notable even by the frothy standards of AI infrastructure funding: Etched was valued at $5 billion in December, then $10.3 billion in July following a $300 million Series C. Doubling that figure again in roughly a month is the kind of move that gets noticed across the entire chip and data center sector.
What makes this round different is who is leading it. Jane Street isn't a traditional venture investor chasing a story - it's a firm that runs massive, latency-sensitive trading workloads and reportedly tested Etched's hardware before buying it outright. In its own statement, Jane Street said it was "pleased with the early results" and cited the precision the chips deliver for demanding workloads, adding that it now has a rack running in its own data center. That's a meaningfully different signal than a typical funding announcement; it suggests a sophisticated technical buyer validated the product under real operational conditions before writing a check. dispensary management software ohio
Two Bottlenecks, Two New Components
Etched's pitch centers on inference - the computation that happens after a user submits a prompt to an AI model - rather than training, which is where most chip competition has historically focused. Co-founder and COO Robert Wachen described inference as splitting into two distinct phases. The "prefill" phase is compute-intensive, requiring the system to process and understand a prompt along with its context. The "decode" phase that follows is memory-intensive, generating the output tokens a user actually sees. Etched built a new chip for the prefill phase that runs at low voltage, which lets it pack in more transistors without the heat problems that plague other high-performance AI chips - allowing it to process more tokens at faster speeds.
For decode, Etched developed what it calls cluster-scale memory, paired with a new interconnect design. Wachen described it as a way to link many chips together so they share a single, very fast, low-latency memory pool. The combined effect, according to the company, is higher throughput at lower cost - a claim that will be tested at scale now that Jane Street's hardware is running in production rather than in a lab.
Shedding an Early Misconception
Etched has spent much of its short life correcting a misunderstanding baked into its own name: that its chips are literally etched to run one specific frontier model, custom-built and inflexible. That was the original design premise, but it isn't how the company operates today. Its systems, delivered as complete "frontier inference clusters" - the company's answer to what Nvidia calls AI factories - are now built to run any frontier model, not a single hardcoded one. That distinction matters commercially, since customers buying inference infrastructure want flexibility as models change rapidly, not a system locked to yesterday's architecture.
A Crowded, High-Stakes Field
Etched's investor roster now includes Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone - a lineup spanning traditional venture capital, sovereign-adjacent growth funds, and now a quant trading firm with direct hardware needs of its own. The Jane Street endorsement is arguably more consequential than the dollar figure attached to it, because it signals demand from a buyer whose entire business depends on shaving milliseconds off computation. Whether that translates into broader adoption against entrenched players remains the open question every AI infrastructure investor is now pricing in.
