Etched, an AI chip startup founded by three Harvard dropouts in 2022, has raised a new $300 million Series C round at a $10.3 billion valuation. The round gives the company a much larger platform for a bet that once looked unusually narrow: building hardware around the needs of modern AI inference.
Co-founder and COO Robert Wachen told TechCrunch that the round was led by Sequoia. Andreessen Horowitz, SK Hynix, Jane Street, and Diffusion Capital also participated, along with other earlier investors.
A valuation jump built on orders and demos
The latest financing marks a sharp rise for Etched. In December, the startup was valued at $5 billion when it raised a $500 million round. In about seven months, that valuation has doubled.
Etched says the new round is the highest valuation ever for a Sequoia-led Series C. That claim lands alongside several operating milestones the company disclosed last month: it had successfully manufactured its homegrown chips, its first full systems were being tested by clients, and it had already booked $1 billion worth of orders.
The company has also attracted a recognizable list of backers. Other investors include Peter Thiel, Andrej Karpathy, Dylan Field, Amjad Masad, and more.
Much of the attention around Etched comes from who has seen the hardware before the broader market has. Access has been limited to investors and early customers, and Wachen said private office demos helped turn well-known AI figures and investors into supporters.
Why Etched was controversial from the start
Etched launched when the idea of building an AI chip specifically for models based on transformer technology was viewed by many as an extreme technical bet. Transformer technology is the architecture behind most modern AI systems, including ChatGPT and Claude.
The company is still pushing back on a common misunderstanding: that its products can run only specific LLMs. Wachen says that is not the case. Etched sells full systems, not just chips, and says those systems can run any AI model.
That includes Mixture of Experts models like DeepSeek and Qwen. In that approach, tasks are split across specialized sub-models rather than relying on one large model. Wachen also says Etched systems can run non-transformer designs like Mamba, which is built on a different underlying architecture known as a state-space model.
The broader market may also be moving closer to Etched's original view. The source notes that the idea of etching parts of a specific AI model directly into silicon to boost performance is no longer seen as far-fetched, with Google reportedly pursuing a similar concept with its Frozen v2 chip for Gemini.
The technical pitch: faster inference
Etched's current claim rests on two components it designed from scratch to accelerate inference. Inference is the computing work that happens after a user submits a prompt.
Wachen describes inference as having two stages: prefill and decode. The prefill phase is where the system understands the prompt and its context. It is mathematically and compute-intensive.
The decode phase is the part that generates output tokens, which become the answer the user sees. That stage requires less computation, but it needs massive amounts of memory.
Etched says its prefill chip runs dramatically faster by operating at a much lower voltage than other AI chips. Wachen calls the approach low-voltage inference. Lower voltage produces less heat, which lets the chip include more transistors.
For decode, the company built a new type of memory and an interconnect technology it calls cluster-scale memory. Wachen says it lets many chips connect and draw on a shared memory pool with very fast, low-latency access.
The promised result is speed at lower cost. For AI companies, that is the core business question behind inference hardware: how quickly a system can respond, how much memory it can access, and how expensive that process becomes at scale.
From garage servers to larger facilities
The founders behind Etched are CEO Gavin Uberti, Robert Wachen, and Chris Zhu. They dropped out of Harvard to start the company, without knowing how difficult fundraising and hiring would become.
Wachen told TechCrunch, "We had no idea how hard it was going to be." He added that the team still has to be humbled by what it will take to actually get to scale.
The early version of Etched was far from the company it is today. Wachen recalled arriving in the Bay Area after telling his parents he was leaving school to start a company, without an office or apartment arranged. He slept on the floor of a friend's unfurnished house and used a towel as a blanket.
The startup also ran chip-design tools from servers set up in an early employee's garage. When the machines needed to be rebooted, Wachen said the employee would call his wife, and she would press the reboot button.
Etched now has 400 people in an office and operates a 2 megawatt data center there. It also just opened a new 80,000 square-foot, 10MW facility in Milpitas, down the road from its main San Jose office.
Despite the new valuation, the company is not claiming the hard part is over. Its rack systems still need to be mass produced and delivered. Wachen said Etched is running tokens in its lab today and working with some of the largest AI companies in the world.
That is the real test behind the funding headline. Etched has turned skepticism into capital, orders, and private demos. Now it has to turn its AI inference systems into scaled hardware customers can use.