Etched’s Series C valuation has landed at $10.3 billion after the AI chip startup closed a $300 million round led by Sequoia, co-founder and COO Robert Wachen confirmed. That is roughly double the $5 billion valuation the company carried in December, achieved in about seven months. Not bad for a firm that was, until recently, best known for the scepticism it attracted.
Andreessen Horowitz, SK Hynix, Jane Street, and Diffusion Capital joined the round alongside earlier backers. The company says this is the highest valuation ever for a Sequoia-led Series C, a claim it has volunteered cheerfully and which, for now, nobody has publicly disputed.
More Investors Than You Might Expect
The investor list has been building quietly for a while. When Etched emerged from stealth, it disclosed over $800 million raised across multiple unannounced rounds prior to this Series C, with earlier backers including Hudson River Trading, Jump Trading, Two Sigma, Ribbit Capital, Radical Ventures, Stanley Druckenmiller, Fei-Fei Li, and Arthur Mensch, among others. Total capital raised is now over $1 billion.
Names like Andrej Karpathy, Geoffrey Hinton, and Peter Thiel also appear on the cap table. Wachen’s explanation for that concentration of AI credibility is straightforward: ‘These are all people who actually tried the hardware and are very excited about it.’ Private demos in the company’s office, apparently, do a lot of the work that press releases cannot.
And if the Wall Street Journal’s reporting is accurate, the fundraising is not done. Etched is reportedly in talks for a separate, larger round at approximately $20 billion valuation, led by existing investor Jane Street. That would be a separate transaction from the $300 million Series C, not a continuation of it, and would roughly quadruple the December valuation in under a year.
Etched Series C Valuation Built on a Specific Technical Bet
The company’s pitch rests on a fairly specific architectural argument. Etched built two custom components targeting the two stages of AI inference: prefill, which processes the incoming prompt, and decode, which generates the output tokens.
For prefill, Etched built a chip that operates at lower voltage than competing AI chips. Lower voltage means less heat, which allows more transistors to be packed in, and Wachen says the result is dramatically faster prefill performance. For decode, the company developed what it calls cluster scale memory: a proprietary interconnect that lets many chips share a single low-latency memory pool at high bandwidth. The claimed outcome is faster inference at lower cost per token.
There is also a persistent misconception Wachen is eager to dispel. The systems are not locked to a single model type. They can run Mixture of Experts architectures like DeepSeek and Qwen, as well as non-transformer designs like Mamba, which uses a state-space model rather than the transformer backbone. The chips are sold as full rack systems, not standalone silicon.
The first batch of chips was manufactured by TSMC, and Etched has reported $1 billion in booked orders. Client testing of the first full systems is under way, though mass production and delivery of rack systems is still ahead.
From a Garage Server Rack to a 10-Megawatt Facility
The company’s operational footprint has grown considerably from its early days, when the founders ran chip-design tools on servers in an employee’s garage, rebooted remotely by his wife when needed. Today, Etched employs 400 people across offices in San Jose and operates a 2-megawatt data centre on site. According to IndexBox, citing EE Times, the company has also opened a separate research and development facility in Milpitas, California, housing a 10-megawatt data centre, a lab, and a quick-turn surface-mount technology line.
For context on the IP side, PitchBook’s company profile shows Etched holds at least four active or pending patents related to tensor operations in AI models, with the first filings dated October 2024.
The founders, CEO Gavin Uberti, Wachen, and CTO Chris Zhu, dropped out of Harvard in 2022 to launch the company before most of the industry had figured out that transformer-specific silicon was worth building. Google is now reportedly pursuing a similar concept with its Frozen v2 chip for Gemini, which suggests the original thesis is not looking quite as wacky as it once did.
‘I think we still have to be humbled by what it will take to actually get to scale,’ Wachen said. Fair enough. But getting to $10.3 billion before the product has fully shipped is, at minimum, a credible start. The next number to watch is whether that reported $20 billion round closes, and whether it does so before the rack systems reach mass production.
