AI chip startup Etched has raised $300 million in a Series C financing at a post-money valuation of $10.3 billion, according to BlockBeats, citing Beating monitoring. The company was founded in 2022 by three Harvard dropouts. Sequoia Capital led the round, with participation from Andreessen Horowitz, SK hynix, Jane Street and Diffusion Capital. Individual investors included Peter Thiel, Andrej Karpathy and Dylan Field.
Etched had previously raised $500 million in December last year at a $5 billion valuation, meaning its valuation doubled in seven months. Sequoia said the deal is the highest-valued Series C investment it has ever led.
Last month, Etched said its first chip had been successfully manufactured by Taiwan Semiconductor Manufacturing Co. Initial full systems are now being tested by customers, and the company said it has already locked in $1 billion in orders. Etched currently has 400 employees and a 2-megawatt data center, and it is conducting joint testing with several leading AI companies around the world.
Built around AI inference rather than standalone chips
Etched’s main pitch is a full-stack system for AI inference, not just chip sales. Co-founder and COO Wachen said inference has two stages: “prefill” and “decode.” The prefill stage is math- and compute-intensive, while decode requires large amounts of memory.
To address that split, the company built a dedicated prefill chip and its own cluster-level shared-memory interconnect technology. Etched said the prefill chip achieves higher transistor density by operating at a much lower voltage than competing products. Its interconnect is designed to let multiple chips access a unified memory pool with very low latency, with the goal of cutting latency while sharply reducing cost.
Wachen responds to model compatibility concerns
Responding to criticism that the chips can only run specific models, Wachen said the system can run any AI model, including mixture-of-experts models such as DeepSeek and Qwen, as well as non-Transformer architectures such as Mamba.
The company has also tried to build credibility by giving in-person demonstrations to investors and early customers. The report said Andrej Karpathy of Anthropic, Noam Brown of OpenAI, and Geoffrey Hinton all gave strong positive feedback after trying the hardware themselves.
Even so, fully configured rack systems are still some distance away from mass production and delivery. Looking back on the company’s early days, Wachen joked that compared with the period when the founders slept on a friend’s floor and used towels as blankets, “now we not only have a mattress, we even have several pillows.” He also said: “When you really believe something is possible and work on it for a long time, you can make it happen.”

