Etched said Tuesday it has booked $1 billion in contract orders for its AI inference systems and raised a total of $800 million, including an unannounced $500 million round that closed in December at a $5 billion post-money valuation. The Nvidia challenger, founded in 2022, said TSMC successfully manufactured its first chip earlier this year, and customer testing of the resulting systems is now underway.
The product Etched is shipping is not a bare chip. The company calls its bundles "frontier inference clusters" — racks that pair its custom silicon with proprietary cooling, networking, and software, all engineered to run frontier-model inference faster, cheaper, and at lower power than Nvidia-based alternatives. Inference, the compute step that happens every time a user submits a prompt, is the largest and fastest-growing cost line for companies serving AI products at scale.
Etched's cap table is unusually dense for a hardware startup at this stage. Investors include VentureTech Alliance, Jane Street, Hudson River Trading, Two Sigma, and Ribbit Capital, with angel checks from Andrej Karpathy, Geoffrey Hinton, Fei-Fei Li, Arthur Mensch, and Scott Wu. Billionaires Stanley Druckenmiller and Peter Thiel are also on the cap table.
Key facts
- 01Etched booked $1B in contract orders for its inference systems before completing customer testing.
- 02An unannounced $500M round closed in December at a $5B post-money valuation, bringing total funding to $800M.
- 03TSMC manufactured Etched's first chip earlier in 2026; the startup was founded in 2022 by Harvard dropouts Gavin Uberti and Robert Wachen.
- 04Investors include Jane Street, Hudson River Trading, Two Sigma, Ribbit Capital, plus angels Andrej Karpathy, Geoffrey Hinton, Fei-Fei Li, and Peter Thiel.
- 05Rival Groq just raised $650M; Cerebras had the year's first breakout AI IPO.
Co-founders Gavin Uberti, the CEO, and Robert Wachen, the president, dropped out of Harvard and became Thiel fellows to start the company. Although Etched framed Tuesday's update as a stealth exit, both founders have been on the record about their chip plans since 2024, when the company had already raised more than $125 million.
The path here was not smooth. On Patrick O'Shaughnessy's Invest Like the Best podcast, Uberti and Wachen said that in 2023, they could not get a single major investor interested, even after circulating a 30-page memo arguing AI would eventually demand specialized silicon rather than general-purpose GPUs. Every firm they pitched passed, and the company spent its early months operating close to cash-out.
The pitch then is the pitch now: Nvidia's GPUs are general-purpose tensor machines, and a chip purpose-built for transformer inference can beat them on the metric customers actually pay for, which is dollars per token served. The $1 billion in advance orders suggests at least some hyperscalers and large model operators believe the math.
The competitive backdrop has flipped from 2023. Cerebras pulled off the year's first breakout AI chip IPO. Groq just raised $650 million. Amazon, Google, and Microsoft each design their own in-house accelerators, and OpenAI announced its first custom chip, built by Broadcom. The thesis Etched's founders could not give away in 2023 — that frontier AI workloads need specialized silicon — is now consensus, with every hyperscaler funding its own version of the answer.
What remains unproven is execution at volume. A booked order is not delivered revenue, and inference clusters live or die on real-world performance per watt against Nvidia's latest generation, not on spec-sheet projections. Etched has yet to publish independent benchmarks comparing its frontier inference clusters to Nvidia configurations on production workloads, and customer testing is the stage where ambitious chip startups historically discover the gap between tape-out and shipping at scale.
The interesting structural question is whether the inference market fragments enough to support a half-dozen specialized challengers, or whether Nvidia's software moat — CUDA, the developer base, the years of optimization work — keeps customers locked in even when a competing chip wins on raw economics. Etched is selling full systems precisely to sidestep that software gravity, betting that a customer who buys the whole rack does not need to care what runs underneath. If the $1 billion in orders converts to repeat business after the first deployments land, the inference layer of the AI stack becomes the most contested chip market since the GPU itself.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
