Etched has closed a $300 million Series C at a $10.3 billion valuation, doubling the AI chip startup's price in the seven months since its December round. Sequoia led the deal, with Andreessen Horowitz, SK Hynix, Jane Street, and Diffusion Capital participating. The company says it is the highest valuation ever attached to a Sequoia-led Series C.
The pace of the mark-up is the headline number. Etched was valued at $5 billion in December on a $500 million raise, meaning the company has doubled inside seven months. That climb is backed by tangible commercial traction: Etched disclosed last month that it had successfully manufactured its first chips at TSMC, that early systems were in customer testing, and that it had booked $1 billion in orders.
The cap table now includes Peter Thiel, Andrej Karpathy, Dylan Field, and Amjad Masad alongside the institutional leads. Founded in 2022 by three Harvard dropouts — CEO Gavin Uberti, COO Robert Wachen, and CTO Chris Zhu — Etched has grown to 400 employees and operates a 2 megawatt data center where it runs inference workloads for what Wachen describes as some of the largest AI companies in the world.
Key facts
- 01Etched closed a $300M Series C at a $10.3B valuation, led by Sequoia with participation from Andreessen Horowitz, SK Hynix, and Jane Street.
- 02The valuation doubled in seven months from the $5B mark set in December's $500M round.
- 03Etched has booked $1B in orders and its first systems, manufactured by TSMC, are in client testing.
- 04The company employs 400 people and runs a 2 megawatt data center for internal testing.
- 05Backers include Peter Thiel, Andrej Karpathy, Dylan Field, and Amjad Masad.
Etched launched around a thesis that most of the industry dismissed: build chips optimized for transformer-based AI models rather than general-purpose accelerators. That bet no longer looks strange. Google is reportedly pursuing a similar approach with its Frozen v2 chip tuned for Gemini.
“Andrej Karpathy from Anthropic, Noam Brown from OpenAI, Geoffrey Hinton, as well as all the investors in the funding round — these are all people who actually tried the hardware and are very excited about it”— Robert Wachen, Etched co-founder and COO
Wachen pushes back on the perception that Etched's systems only run specific large language models. The systems can run any AI model, he says, including Mixture of Experts architectures such as DeepSeek and Qwen, and non-transformer designs like Mamba, which is built on state-space methods rather than attention.
The technical claim rests on two custom components aimed at inference — the compute step that happens after a user submits a prompt. Wachen splits inference into a prefill phase, which parses the prompt and is compute-heavy, and a decode phase, which generates output tokens and is memory-heavy. Etched built a prefill chip that runs at what Wachen calls low-voltage inference, generating less heat and allowing more transistors per die. For decode, the company designed a shared memory pool it calls cluster scale memory, letting many chips draw from a common memory at low latency.
Access to the hardware has been tight, which is part of why skepticism has lingered. Investors and a small group of early customers have seen private demos. That gated approach is also how Etched recruited the technical names on its investor list — Karpathy, OpenAI's Noam Brown, and Geoffrey Hinton all tried the hardware before backing it, according to Wachen.
The founders' path was not clean. Wachen recalls landing in the Bay Area with no office and no apartment, sleeping on the floor of a friend's unfurnished house with a towel for a blanket. Early chip-design servers ran in an employee's garage, and when the rig needed a reboot, the employee's wife would go hit the button.
The remaining risk is execution. Etched still has to move from lab-scale token generation and customer pilots to mass production and delivery of rack systems at the scale its $1 billion order book implies. Data center customers do not tolerate slipped delivery dates, and the AI accelerator market is now crowded with well-capitalized competitors — Nvidia's dominant position, AMD's expanding footprint via its recent Anthropic deal, and hyperscaler in-house silicon programs at Google, Amazon, and Microsoft. Etched's inference-only pitch is sharper than a general-purpose GPU, but sharper pitches are also easier to route around if the delivered performance does not match the demo.
For the AI chip market, the Etched round is a data point that inference-specific silicon is now a fundable category on its own, separate from the training-focused GPU race. A $10.3 billion valuation for a company still shipping its first systems tells you that Sequoia and SK Hynix — the latter one of the world's largest memory suppliers — think the economics of running deployed models at scale are diverging enough from training that a purpose-built stack can carve out room next to Nvidia. Whether Etched can hold that room depends on the next twelve months of actual deliveries, not on the funding round.
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.
