Etched’s $300 Million Raise Doubles Its Valuation to $10.3 Billion

Etched, the AI chip startup building hardware specifically for running AI models rather than training them, has closed a $300 million Series C funding round at a $10.3 billion valuation, co-founder and COO Robert Wachen told TechCrunch. The round was led by Sequoia Capital, with Andreessen Horowitz, SK Hynix, Jane Street, and Diffusion Capital also participating alongside earlier investors.

A Valuation That Doubled in Seven Months

Sequoia, which led the round, said in its own announcement that it is “delighted to be partnering with Etched and leading their $300 Million Series C at a $10 Billion pre-money valuation.” Adding the new capital brings the post-money figure to the $10.3 billion Wachen gave TechCrunch. Etched had been valued at $5 billion just seven months earlier, in December, when it raised a $500 million round, according to the company. Etched says the new valuation is the highest ever for a Sequoia-led Series C. Backers beyond the lead investors include Peter Thiel, Andrej Karpathy, Dylan Field, and Amjad Masad, TechCrunch reported.

Betting Silicon Against General-Purpose GPUs

Founded in 2022 by three Harvard dropouts, CEO Gavin Uberti, COO Robert Wachen, and CTO Chris Zhu, Etched bet early that transformer-based models (the architecture behind most modern AI systems, including ChatGPT and Claude) were common enough to justify chips built specifically for that workload, instead of general-purpose GPUs. TechCrunch notes that when Etched launched, the idea was “considered wild if not wacky.” The company still faces skeptics, Wachen said, in part because access to its systems has so far been limited to investors and early customers who saw the hardware through private demos in Etched’s office.

Etched came out of stealth on June 30, revealing in a press release that it had already raised $800 million across previously undisclosed rounds, including the $500 million December round, and had signed more than $1 billion in customer contracts. Combined with the new Series C, Etched’s disclosed funding now totals roughly $1.1 billion. The company said it achieved “first-pass (A0) silicon success” on TSMC’s N4P process in under three years from its seed round, and its inference systems are already running models including DeepSeek, Qwen, Mamba, and Llama for early customers.

Splitting Prefill From Decode

Etched’s technical pitch centers on treating the two stages of AI inference, the computing that happens after a user submits a prompt, as separate hardware problems. “Inference is built in two stages,” Wachen told TechCrunch, “prefill and decode.” The prefill phase reads and interprets the prompt and its context, a compute-intensive step. The decode phase generates the output tokens the user actually sees: a step that needs far less computation but far more memory.

For prefill, Wachen said Etched built a chip that runs “at a much lower voltage than any other AI chip,” a technique the company calls low-voltage inference. Lower voltage generates less heat, which lets the chip pack in more transistors. For decode, the company designed a new memory and interconnect approach it calls cluster-scale memory, which Wachen described as letting many chips “connect together and use a shared memory pool at a very, very fast, low latency.” The combination, he said, delivers higher speeds at lower cost than general-purpose GPUs.

Not Locked to a Single Model

Because Etched sells complete systems rather than standalone chips, Wachen said the company still fights a misconception that its hardware is designed to run only specific large language models. He told TechCrunch the systems support any AI model, including Mixture of Experts designs like DeepSeek and Qwen, which split tasks across specialized sub-models rather than relying on one large model, as well as non-transformer architectures like Mamba, a state-space model. TechCrunch separately noted that Google is reportedly pursuing a similar concept with a chip called Frozen v2 for Gemini, suggesting the idea of hardware tuned to a specific model family is gaining traction beyond Etched.

From a Friend’s Floor to a Second Data Center

Wachen’s account of Etched’s early days is a reminder of how recent the company’s rise has been. After telling his parents he was leaving Harvard to start the company, he said he arrived in the Bay Area with no office or apartment arranged and slept on the floor of a friend’s unfurnished house, “using a towel as a blanket,” he told TechCrunch. The founders ran their chip-design tools on servers set up in an early employee’s garage, and rebooting them meant a phone call asking his wife to walk over and press the button.

Etched now employs more than 400 people, according to the GlobeNewswire release, with most hires coming from Nvidia, Broadcom, Google’s TPU program, SK Hynix, and high-frequency trading firms. “We’ve long believed inference would be one of the defining trends in AI,” said Ken Fox, founder and partner at investor Stripes, in the release. The company operates a 2-megawatt data center at its San Jose office and recently opened an 80,000-square-foot, 10-megawatt facility in Milpitas, with a stated path to gigawatt-scale deployment in 2027. “We’re running tokens in our lab today, working with some of the largest AI companies in the world,” Wachen told TechCrunch.