Etched, an AI chip startup founded in 2022 by three Harvard dropouts, has closed a $300 million Series C funding round at a valuation of $10.3 billion, co-founder and chief operating officer Robert Wachen told TechCrunch.
The round was led by Sequoia, with participation from Andreessen Horowitz, SK Hynix, Jane Street, and Diffusion Capital, as well as previous investors. Other backers of the company include names like Peter Thiel, Andrei Karpathy, Dylan Field, and Amjad Massad.
Etched was valued at $5 billion when it raised $500 million in December, which means its valuation has doubled in about seven months. The company says this is the highest valuation ever for a Sequoia-led Series C. Last month, Etched announced that it had successfully manufactured its own chip, that the first complete system was being tested by customers, and that it had already booked orders worth $1 billion.
Etched was launched at a time when the idea of building dedicated chips for AI models based on Transformer technology (the architecture behind most modern AI systems, including ChatGPT and Claude) was considered far-fetched, if not far-fetched. The company is still battling the perception that its products, which are sold as complete systems and not just chips, include chips designed to perform only specific LLMs.
Not so, Wachen explains. The system can run any AI model, including expert mixture models such as DeepSeek and Qwen (architectures that divide tasks among specialized submodels rather than relying on one large model). You can also run non-transformer designs like Mamba, which are built on a different underlying architecture known as a state-space model. (Interestingly, the idea of etching parts of certain AI models directly into silicon to improve performance is no longer considered far-fetched; Google is reportedly pursuing the same concept with its Frozen v2 chips for Gemini.)
Still, Etched’s claim to fame today is that it designed two new components from the ground up to speed up inference, the computing process that runs after a user submits a prompt.
“Inference is built in two stages: prefill and decode,” Wachen says. The “prefill phase” involves understanding the prompt, including its context. It is mathematical and computationally intensive. The “decode” phase produces an output token (the actual answer that is displayed to the user). It requires less computation, but requires more memory.
Etched has created a pre-filled chip that operates “dramatically” faster by operating at a much lower voltage than other AI chips. We call this low voltage reasoning. ” he promises. Lower voltages generate less heat and allow more transistors to be placed on the chip.
For the decoding process, Etched has created a new type of memory and “an interconnect technology called cluster-scale memory, which allows many chips to connect together and use a shared memory pool at very high speeds and low latency,” he says. The result, Etched promises, is both faster and lower costs.
Because the startup was launched before most of the tech industry (other than Nvidia) understood the specialized computing needs of AI, its founders, CEO Gavin Uberti, Wachen, and CTO Chris Zhu, have faced many skeptics who continued to have doubts even after the company announced its first batch of silicon was successfully manufactured by TSMC.
Much of this is due to the fact that so few people have access to the system. So far, access has been limited to investors and early customers. In fact, Etched first gained a list of big-name investors by showing private demos in its offices.
“Andrej Karpathy from Anthropic, Noam Brown from OpenAI, Geoffrey Hinton, and all the investors in the funding round, these are all people who have actually tried the hardware and are really excited about it,” Wachen said.
Still, there is still a long and difficult road ahead before racking systems can be mass-produced and delivered. The trio famously dropped out of Harvard to start Etched, but at the time they didn’t know how to raise money (much less how much money they needed) or how to hire.
“We had no idea how difficult it was going to be,” Wachen says. “I think we still need to be humble about what it will take to actually scale.”
Wachen remembers telling his parents he was quitting school to start a startup and landing in the Bay Area without an office or an apartment. He was sleeping on the floor of a friend’s unfurnished house.
“I remember staying at a friend’s house that was going to be sold and using towels as blankets,” he laughs. Eventually, the founders installed the servers needed to run the chip design tools in an early employee’s garage, and “every time I needed a reboot, I would call my wife and she would go and press the reboot button.”
Today, 400 people are busy working in the office, where Etched operates a 2-megawatt data center. And the company just opened a new 80,000-square-foot, 10MW facility in Milpitas, just a short distance from its San Jose headquarters.
“We are currently working with some of the largest AI companies in the world to run tokens in their labs,” he says.
Wachen now has blankets and a mattress, “and a pillow. I have more than one pillow,” he jokes.
More importantly, he and his co-founders never let the doubters stop them. “We’ve come a long way. It’s a very different world. But I think if you really think something is possible and you work at it for a long time, you can make it happen.”
Note: This article has been updated to include reference to the new 10MW facility.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
