Etched secures $700M round at a $21 billion valuation
AI hardware startup Etched has raised a further $700 million at a $21 billion valuation in a funding round led by quantitative trading firm Jane Street. The valuation has nearly doubled in a month from the $10.3 billion level attached to Etched’s $300 million Series C in July, and follows a $5 billion valuation in December.
Jane Street tested Etched’s AI hardware before investing and purchasing it. In its announcement, the firm said it was pleased with early chip results and had installed its own rack in its datacentre to support demanding workloads.
Inference systems are the product
Etched supplies complete systems it calls “frontier inference clusters”. These systems are intended for inference, the computation that takes place after a user submits a prompt to an AI model. Co-founder and COO Robert Wachen described inference as two stages: prefill and decode.
During prefill, a system processes the prompt and its context, making the stage mathematically and computationally intensive. Decode generates the output tokens that form the answer seen by the user, and is memory-intensive. Etched says it has designed separate components to address those different workloads.
New chip, memory and interconnect design
For prefill, the company developed a low-voltage chip designed to accommodate more transistors without the heat issues associated with other high-end AI chips. Etched says this enables it to process more tokens more quickly.
For decode, Etched developed a new memory approach and interconnect called cluster-scale memory. The company says it allows many chips to connect to a shared memory pool with high speed and low latency. Etched presents the combined design as a route to higher speeds and lower costs.
The funding follows Etched’s $300 million Series C funding and marks another rapid step in Etched’s valuation. The company is also seeking to correct an impression from its early development that its chips were designed around a single frontier model; it now says its systems can run any frontier model.
What the round signals for buyers
Etched’s investor group includes Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Tiger Global, Bain Capital Ventures and Blackstone, alongside Jane Street. For businesses assessing AI infrastructure, the relevant question is how a complete inference system performs on their own prefill and decode workloads, including latency, memory behaviour and cost, rather than valuation alone.