Etched's AI chip valuation doubles to $21 billion in a single month

Etched, a San Jose AI chip startup founded by three Harvard dropouts, has seen its valuation double to $21 billion in under a month after raising $700 million in a new round led by Jane Street, which also became the first customer for its rack-scale AI inference system. The company, whose Sohu ASIC is designed specifically for transformer inference, previously raised $300 million at a $10.3 billion valuation on July 23 in a Series C led by Sequoia Capital, and claims over $1 billion in booked orders.

Etched has gone from a $10.3 billion valuation to $21 billion in less than a month because Jane Street is no longer just an investor. It is a customer with the startup's hardware in hand. Three years ago, Etched was three Harvard dropouts with a Thiel Fellowship and a bet that Nvidia's chips were the wrong tool for running deployed AI models. Today it's worth $21 billion. The San Jose startup has raised $700 million in a new round led by Jane Street, according to The Wall Street Journal, after the quantitative trading firm became the first customer for Etched's rack-scale AI inference system. That is the part you should watch. AI chip startups can raise money on drawings and benchmarks. Only for so long. A customer taking delivery is a different signal. The speed of the valuation jump is still hard to ignore. Etched said on July 23 that it had raised $300 million at a $10.3 billion valuation in a Series C led by Sequoia Capital, with Andreessen Horowitz and Jane Street joining the round, alongside Diffusion and SK Hynix. TechCrunch reported in June that the company had been valued at $5 billion in a December 2025 round. Now Jane Street is leading another check. The number attached to Etched has doubled in under a month. The narrow bet is the point Etched's chip, Sohu, is not trying to be an all-purpose GPU. It is an ASIC built for transformer inference, the work that happens after a model has already been trained and starts answering prompts. Nvidia sells chips that can handle gaming and simulation, training and inference. Sohu does one job. That is the bet. Etched raises $300M at a $10.3B valuation on a bet that transformer-only chips will displace Nvidia at the inference layer https://startupfortune.com/etched-raises-300m-at-a-103b-valuation-on-a-bet-that-transformer-only-chips-will-displace-nvidia-at-the-inference-layer/ Etched, founded by three Harvard dropouts, closed a $300M Series C at a $10.3B valuation on July 23, led by Sequoia in what the firm calls its highest-valued Series C ever. The company's Sohu ASIC hardcodes the transformer architecture into silicon, claiming a 20x inference speed advantage over Nvidia's H100, and has already booked over $1 billion... - etched AI chip funding announcement https://startupfortune.com/etched-raises-300m-at-a-103b-valuation-on-a-bet-that-transformer-only-chips-will-displace-nvidia-at-the-inference-layer/ - transformer inference chip replaces Nvidia https://startupfortune.com/etched-raises-300m-at-a-103b-valuation-on-a-bet-that-transformer-only-chips-will-displace-nvidia-at-the-inference-layer/ Gavin Uberti, Etched's co-founder and CEO, told TechCrunch in 2024 that one Sohu server could replace 160 Nvidia H100 GPUs on transformer workloads, and TechCrunch reported that the chip is made on TSMC's 4-nanometer process. Those are company claims, not independent market proof, but they explain why investors are crowding in. If the architecture works at scale, Etched is not selling a slightly cheaper chip. It is selling a narrower machine for the most expensive part of running AI products. The company has moved beyond a single chip story. Etched now talks about frontier inference clusters, full rack-scale systems where chips, boards and cooling sit alongside software and memory, all designed together. On its own site, Etched said its A0 silicon came back from TSMC N4P earlier this year, that it had opened a 10-megawatt lab near its San Jose office, and that first racks were due to ship this summer. The Journal reported that Etched had chips running inference workloads 44 days after receiving test silicon from TSMC, a timeline the company says usually takes six months or more. The machine matters. The valuation is now the test Etched says it has booked more than $1 billion in orders. That is real traction for a chip company that was still describing itself as coming out of stealth earlier this summer, but it is also a small number beside a $21 billion valuation. You do not need to be cynical to see the gap. You only need to know how hard hardware is. There is some substance under the hype. The Journal reported that Etched has about 400 employees and that roughly 15% of them previously worked at Nvidia. The investor list now includes Jane Street, Sequoia, Andreessen Horowitz, Kleiner Perkins, Tiger Global, Bain Capital Ventures, Blackstone and Peter Thiel. Uberti started the company with Chris Zhu and Robert Wachen after leaving Harvard, and the three became Thiel Fellows. This is no longer a two-person chip dream with a clever deck. Still, a $700 million round does not make Nvidia less dangerous. Nvidia has the software stack and the customer relationships. It also has the supply chain muscle - and the habit of moving faster than its challengers expect. Groq, Cerebras and the cloud giants have all chased pieces of the same inference opportunity. Investors clearly believe there is room for chips that do less, better. And it should. The cost of serving AI models is becoming too large for every customer to accept a general-purpose answer forever. Look, Jane Street installing Etched's hardware is a stronger signal than most funding announcements offer. But the next proof will not come from another valuation. It will come from more racks running in more data centers, with customers willing to keep paying after the first test works. Etched bets $800 million that transformer silicon will outlast the GPU era https://startupfortune.com/etched-bets-800-million-that-transformer-silicon-will-outlast-the-gpu-era/ Etched emerged from stealth on June 30, 2026, announcing $800 million raised, a working transformer-specific chip called Sohu, and over $1 billion in signed customer contracts. The startup's bet is that hardcoding the transformer architecture into silicon delivers inference performance that general-purpose Nvidia GPUs cannot match, with early... - transformer inference chip custom silicon https://startupfortune.com/etched-bets-800-million-that-transformer-silicon-will-outlast-the-gpu-era/ - Etched AI chip TSMC production https://startupfortune.com/etched-bets-800-million-that-transformer-silicon-will-outlast-the-gpu-era/ Also read: Anthropic Shows AI Agents Can Infect Each Other With a Self-Spreading Goal https://startupfortune.com/anthropic-shows-ai-agents-can-infect-each-other-with-a-self-spreading-goal/ • Claude AI Suffers Widespread Outage Across All Its Models on August 18 https://startupfortune.com/claude-ai-suffers-widespread-outage-across-all-its-models-on-august-18/ • AI Is Creating More Jobs Than It Cuts in the UK, Lloyds Finds https://startupfortune.com/ai-is-creating-more-jobs-than-it-cuts-in-the-uk-lloyds-finds/