Etched reaches $10.3B valuation with custom AI inference chips

Etched has closed a $300 million Series C round at a $10.3 billion valuation, doubling its worth since December and marking the highest valuation ever for a Sequoia-led Series C, according to co-founder and COO Robert Wachen.

Sequoia led the round, with participation from Andreessen Horowitz, SK Hynix, Jane Street, and Diffusion Capital. The AI chip startup, founded in 2022 by three Harvard dropouts, has also attracted backing from Peter Thiel, Andrej Karpathy, Dylan Field, and Amjad Masad. Last month, Etched announced successful manufacturing of its chips through TSMC and reported $1 billion in booked orders, with first systems now in client testing.

Why it matters

Etched's rapid valuation growth reflects investor confidence in specialized inference hardware as AI deployment costs become a critical bottleneck for enterprises. The company's approach—designing chips optimized for the distinct computational demands of prefill and decode stages—addresses a specific pain point that general-purpose accelerators handle less efficiently. If Etched delivers on its performance and cost promises at scale, it could reshape infrastructure economics for companies running large language models in production.

The technical architecture

The startup designed two novel components targeting different phases of inference—the computing process that occurs after a user submits a prompt. The prefill phase analyzes the prompt and context, requiring intensive computation. The decode phase generates output tokens with less computation but massive memory demands.

For prefill, Etched built a chip operating at much lower voltage than competing AI accelerators. "Lower voltage generates less heat, which allows the chip to pack in more transistors," Wachen explained. For decode, the company developed what it calls cluster-scale memory—an interconnect technology enabling multiple chips to share a memory pool at low latency.

Contrary to early perceptions, Etched's systems run any AI model architecture, including Mixture of Experts designs like DeepSeek and Qwen, as well as non-transformer models like Mamba. The company sells complete systems rather than standalone chips.

From garage servers to 400 employees

The founding team—CEO Gavin Uberti, Wachen, and CTO Chris Zhu—launched Etched when building transformer-specific chips was considered impractical. They faced persistent skepticism even after announcing successful silicon manufacturing.

Wachen recalled arriving in the Bay Area after leaving Harvard with no office or housing arranged, sleeping on a friend's floor using a towel as a blanket. The team initially ran chip-design servers in an early employee's garage, relying on his wife to manually reboot systems when needed.

Etched has since grown to 400 employees and operates a 2-megawatt data center. The company secured investor backing by offering private hardware demonstrations in its office—a strategy that brought in backers including Karpathy from Anthropic, Noam Brown from OpenAI, and Geoffrey Hinton.

"When you really think something's possible, and you just work at it for a long time, you can do it," Wachen said.

The details were first reported by TechCrunch.