AI chip startup Etched defies skeptics, hits $10.3B valuation from big-name investors

Etched, the AI chip startup founded by three Harvard dropouts in 2022, has closed a $300 million Series C funding round at a $10.3 billion valuation, co-founder and COO Robert Wachen tells TechCrunch.

The round was led by Sequoia, with Andreessen Horowitz, SK Hynix, Jane Street, and Diffusion Capital also participating, along with other, earlier investors. Other backers of the company include names like Peter Thiel, Andrej Karpathy, Dylan Field, Amjad Masad, and more.

Etched was previously valued at $5 billion in December when it raised a $500 million round, meaning it has doubled its valuation 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 homegrown chips , that its first full systems were being tested by clients, and it had already booked $1 billion worth of orders.

Etched launched at a time when the idea of building a chip specifically for AI models based on transformer technology (the architecture behind most modern AI systems, including ChatGPT and Claude) was considered wild if not wacky. The company is still battling the perception that its products — which are sold as full systems, not just chips — involve chips designed to run only specific LLMs.

That’s not the case, Wachen explains. The systems can run any AI model, including Mixture of Experts models like DeepSeek and Qwen — an architecture that splits tasks across specialized sub-models rather than relying on one large model — as well as non-transformer designs like Mamba, which is built on a different underlying architecture known as a state-space model. (Interestingly, the idea of etching parts of a specific AI model directly into silicon to boost performance isn’t considered far-fetched anymore. Google is reportedly pursuing the same concept with its Frozen v2 chip for Gemini.)

Still, Etched’s claim to fame today is that it designed two new components from scratch to speed up inference — the computing process that happens after a user submits a prompt.

“Inference is built in two stages,” Wachen says, “prefill and decode.” The “prefill phase” involves understanding the prompt, including context. It’s mathematically and compute-intensive. The “decode” phase generates the output tokens (the actual answer the user sees). It requires less computation but needs massive amounts of memory.

Etched created a prefill chip that operates “dramatically” faster, he promises, “by running at a much lower voltage than any other AI chip. We call this low-voltage inference.” Lower voltage generates less heat, which allows the chip to pack in more transistors.

For the decode process, Etched created a new type of memory and “interconnect technology that we call cluster scale memory. It allows many chips to connect together and use a shared memory pool at a very, very fast, low latency,” he says. The result, Etched promises, is high speeds but lower costs.