Etched raises $700M at a $21B valuation in Jane Street-led round

Etched, an AI inference systems startup founded by Gavin Uberti, Robert Wachen, and Chris Zhu, raised $700 million at a $21 billion valuation in a round led by Jane Street, which tested Etched's hardware and received its first rack in its data center. The company has raised $1.9 billion total and secured over $1 billion in customer contracts, with participation from Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Tiger Global, Peter Thiel, Bain Capital Ventures, Neo, Stripes, Primary Venture Partners, Positive Sum, Diffusion, Argo, and Blackstone. The funding follows a $300 million Series C at a $10.3 billion valuation reported on July 23, roughly doubling Etched's valuation in less than four weeks.

Etched https://www.etched.com/?ref=runtimewire has raised $700M at a $21B valuation in a round led by Jane Street, Reuters reported August 18 https://www.reuters.com/technology/ai-chip-startup-etched-valued-21-billion-latest-funding-round-2026-08-18/?ref=runtimewire . The AI inference systems startup was founded by Gavin Uberti https://www.semi.org/en/bio-gavin-uberti?ref=runtimewire , Robert Wachen @robertwachen https://x.com/robertwachen?ref=runtimewire and Chris Zhu @czhu1729 https://x.com/czhu1729?ref=runtimewire . Jane Street https://www.janestreet.com/?ref=runtimewire led the round after testing Etched's hardware, according to Etched's August 18 announcement https://www.etched.com/progress/from-zero-to-one?ref=runtimewire . Etched said the trading firm received its first rack https://www.etched.com/progress/from-zero-to-one?ref=runtimewire , publicly identifying Jane Street as a customer with the system in its own data center. Kleiner Perkins https://www.kleinerperkins.com/?ref=runtimewire , Sequoia Capital https://sequoiacap.com/?ref=runtimewire , Andreessen Horowitz https://a16z.com/?ref=runtimewire and Tiger Global https://www.tigerglobal.com/?ref=runtimewire participated, alongside Peter Thiel, Bain Capital Ventures, Neo, Stripes, Primary Venture Partners, Positive Sum, Diffusion, Argo and Blackstone. Etched says it has raised $1.9B in total and secured more than $1B in customer contracts across public and private AI companies and cloud providers. Reuters reported both figures https://www.reuters.com/technology/ai-chip-startup-etched-valued-21-billion-latest-funding-round-2026-08-18/?ref=runtimewire . The funding follows a $300M Series C at a $10.3B valuation reported on July 23 https://techcrunch.com/2026/07/23/ai-chip-startup-etched-defies-skeptics-hits-10-3b-valuation-from-big-name-investors/?ref=runtimewire . In less than four weeks, Etched has roughly doubled its valuation and raised another $700M. Etched says the new capital will finance production https://www.etched.com/progress/from-zero-to-one?ref=runtimewire , putting a $21B price on the founders' ability to move from working silicon and customer tests into repeatable manufacturing. A dorm-room thesis reaches the data center Uberti and Zhu started Etched as a dorm-room project while studying mathematics and computer science at Harvard. They closed a $5.36M seed round in 2023, left school and moved to the Bay Area, according to an early profile of the founders https://www.canopy.space/members/member-profile-etched/?ref=runtimewire . The founders' wager was direct: transformer models would become important enough to justify hardware designed around their workloads. At the time, that meant rejecting the flexibility of general-purpose graphics processors and betting that model architecture would remain stable long enough for specialized silicon to win on speed and cost. Uberti had worked on AI compilers at OctoML, writing matrix-multiplication kernels for the Apache TVM compiler and contributing to MLPerf Tiny work. Zhu's background combined mathematics with high-performance computing and combinatorics research. Wachen, Etched's president, took responsibility for fundraising, operations and customer development as the project became a semiconductor business. The early operation bore little resemblance to a company now valued at $21B. Wachen told TechCrunch in July https://techcrunch.com/2026/07/23/ai-chip-startup-etched-defies-skeptics-hits-10-3b-valuation-from-big-name-investors/?ref=runtimewire that he arrived in the Bay Area without an apartment and slept on the floor of a friend's unfurnished home, using a towel as a blanket. Etched initially ran chip-design servers from an early employee's garage. "We had no idea how hard it was going to be," Wachen said in that interview. Etched says it employs more than 400 engineers https://www.etched.com/?ref=runtimewire . TechCrunch reported that Etched operates a 2-megawatt data center at its San Jose office and opened an 80,000-square-foot, 10-megawatt facility in nearby Milpitas. The scale-up has happened before Etched has published revenue, shipment volume or independently tested production benchmarks. Jane Street becomes the proof point Jane Street's role gives the announcement more weight than another investor endorsement. Etched says Jane Street tested the chip, led the round and installed the first rack in its data center https://www.etched.com/progress/from-zero-to-one?ref=runtimewire . Those are company-reported customer findings rather than a public benchmark, but they indicate that Etched has moved beyond simulations and private laboratory demonstrations. Reuters did not identify customers in its report on the financing. Etched's own announcement identifies Jane Street, while the company has not published the mix of deposits, purchase commitments or delivery conditions behind its claim of more than $1B in customer contracts. The financing figures also leave a small gap. Adding the $300M July round and Tuesday's $700M financing produces $1.8B, about $100M below the cumulative figure reported to Reuters. The difference could reflect an additional tranche or rounding across earlier financings. Etched is selling the rack, not merely the chip Etched originally introduced Sohu as an application-specific integrated circuit built around transformer inference. The company now calls its systems "frontier inference clusters" https://www.etched.com/?ref=runtimewire and says it co-designs chips, racks, software and manufacturing methods. That change reflects the practical difficulty of competing in data centers. A fast processor has limited value if customers must assemble the rest of the system themselves. Etched is taking responsibility for the full rack so it can tune power, memory movement and software around the silicon, while giving buyers a deployable alternative to general-purpose AI systems. Etched says its architecture uses low-voltage inference to increase compute density for the prefill stage, when a model processes a prompt and its context. For token generation, Etched says its cluster-scale memory design connects chips to a shared memory pool with lower latency. Those performance claims remain company assertions, with wider customer access and public testing still limited. In its June 30 announcement, Etched said its first A0 silicon came back from TSMC's N4P process and that it had kicked off production to fulfill more than $1B in customer contracts https://www.etched.com/progress/frontier-inference-clusters?ref=runtimewire . The contract figure has not been independently substantiated, and Etched has not disclosed production shipment volume. The founders now have financial backing that few young hardware companies ever reach. The $21B valuation sets a harder operating benchmark alongside it. Etched must manufacture racks reliably, deliver them on schedule and prove that specialization can remain useful as model architectures change. The Jane Street installation is an early deployment. Etched's next task is to make such installations routine.