Etched, a startup building AI chips designed only to run inference, disclosed on Tuesday that it has booked $1 billion in contract orders for systems powered by its silicon. TechCrunch reported the figures from the company’s own progress update, issued after TSMC completed manufacturing of Etched’s chip earlier this year. The stakes are straightforward: inference, the process of running a trained model to answer a query, is now the largest recurring cost for AI companies serving customers at scale, and Etched is betting that a chip built for nothing else can beat Nvidia’s general-purpose GPUs on that job alone.
The company calls its product a “frontier inference cluster.” It bundles the chips with custom racks and software, aimed at running frontier models faster, more cheaply, and with better power efficiency than competing hardware, according to Etched. Those are the company’s own performance claims. TechCrunch’s report does not include independent benchmark results comparing Etched’s systems against Nvidia’s current lineup, so the efficiency claims remain unverified by a third party.
Etched also disclosed that it has now raised $800 million in total funding. The most recent round, $500 million, closed in December at a $5 billion post-money valuation and had not been previously announced. Stripes led that round, alongside VentureTech Alliance, Jane Street, Hudson River Trading, Two Sigma, and Ribbit Capital. The investor list also includes individual backers with direct AI research credibility: Andrej Karpathy, Geoffrey Hinton, Fei-Fei Li, Arthur Mensch, and Scott Wu, plus billionaires Stanley Druckenmiller and Peter Thiel.
Etched frames Tuesday’s news as coming out of stealth, but that framing understates the company’s history. Founders Gavin Uberti and Robert Wachen, who left Harvard as Thiel Fellows to start the company in 2022, have discussed their chip strategy with TechCrunch publicly since 2024, and had already raised more than $125 million by that point. On Patrick O’Shaughnessy’s “Invest Like the Best” podcast, the founders described a starkly different fundraising climate in 2023: a 30-page memo arguing that AI workloads would eventually demand specialized silicon rather than general-purpose GPUs, pitched to every major investor they could reach. All of them passed, and the company was reportedly operating close to month-to-month.
That contrast matters more than the valuation headline. A specialized-chip thesis that could not find a single believer in 2023 now commands $5 billion and attracts Thiel-and-Druckenmiller-tier capital, without Etched having shipped a commercial product beyond ongoing customer testing. The shift says less about Etched’s technology, which has not been independently benchmarked, than about how thoroughly investor appetite for inference alternatives to Nvidia has changed in three years.
Etched is not alone in that shift. Cerebras completed the year’s first breakout AI chip IPO, and Groq just raised $650 million. Amazon, Google, and Microsoft each build in-house AI chips for their own clouds, and OpenAI recently announced its first custom chip, built with Broadcom. Nvidia still controls the overwhelming majority of AI training and inference hardware, but the number of well-funded challengers targeting inference specifically has multiplied in a single year.
For enterprise buyers evaluating 2027 inference contracts, Etched’s $1 billion order book is a signal worth tracking, not a benchmark worth trusting yet. The company has not disclosed customer names, contract terms, or independent performance data, so any procurement decision based on Etched’s claims should wait for those numbers or for results from the customers currently testing the hardware.
Reported by TechCrunch on June 30, 2026.
