Etched raises $700 million as valuation doubles to $21 billion in a month
Etched announced on Tuesday that it raised $700 million at a $21 billion valuation, in a round led by Jane Street. The quant trading firm tested and bought Etched's AI hardware before leading the investment. The jump is fast even by AI standards: Etched was valued at $5 billion in December, then raised a $300 million Series C at a $10.3 billion valuation in July. The new round doubles that figure to $21 billion, an increase of nearly $11 billion in a single month.
Etched sells its technology as complete systems it calls "frontier inference clusters," positioned against what Nvidia calls its "AI factories." Co-founder and COO Robert Wachen told TechCrunch that investor enthusiasm comes from two components Etched built from scratch to speed up inference, the computing that happens after a user submits a prompt. Wachen described inference as built in two stages: a compute-intensive "prefill" phase, where the system parses the prompt and its context, and a memory-intensive "decode" phase, where the system generates the output tokens the user actually sees. Etched built a prefill chip that runs at low voltage, letting it pack in more transistors without the heat problems that affect other high-end AI chips, so it can process more tokens faster. For decode, Etched built a new type of memory and interconnect it calls cluster-scale memory. Wachen said it lets many chips connect and share a memory pool at very fast, low latency, which Etched says translates into higher speeds and lower costs.
The company is also still shedding an early misconception baked into its name: that each chip is custom-etched to run one specific frontier model. That was the original plan, but Etched says it no longer holds, and its systems can now run any frontier model.
In a blog post announcing the round, Jane Street wrote: "We tested the chip and are pleased with the early results. Etched's unique approach to inference delivers the precision we will need to support our most demanding workloads. We're excited to now have our own rack running in our datacenter." Etched's other investors include Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo and Blackstone.
Key facts
- Etched raised $700 million led by Jane Street at a $21 billion valuation, announced Tuesday.
- The valuation doubled from $10.3 billion (July Series C) to $21 billion in about a month, up nearly $11 billion; it stood at $5 billion in December.
- Jane Street tested and bought Etched's hardware before leading the round, and says it now runs its own rack in its datacenter.
- Etched's systems split inference into a prefill phase (a low-voltage chip built for more transistors and fewer heat problems) and a decode phase (a new cluster-scale memory and interconnect for shared, low-latency memory pooling).
- Etched no longer builds chips locked to one frontier model, despite what its name implies; its systems now run any frontier model.
Why it matters
The speed of the step-up, from a $5 billion valuation in December to $21 billion now, signals strong investor appetite for inference-specialized hardware as an alternative to general-purpose GPUs. Jane Street's involvement is notable because the fund became a customer first, testing and buying Etched's hardware before leading the funding round, and then wrote publicly about running its own rack. That is a stronger signal of product validation than a typical venture check, since a quant trading firm has an unusually low tolerance for latency and precision problems.
Who it affects
AI infrastructure buyers weighing inference hardware beyond Nvidia's GPUs, cloud and datacenter operators evaluating specialized inference clusters, and Etched's existing and new investors, including Jane Street, Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo and Blackstone. Nvidia is named directly as the competitor whose "AI factories" framing Etched is positioning against.
How to use it
Etched sells its technology as full systems, "frontier inference clusters," rather than standalone chips. The architecture separates the two stages of inference: a low-voltage prefill chip handles prompt understanding and context, packing in more transistors without the heat penalties of other high-end AI chips, while a separate cluster-scale memory and interconnect handles the decode stage, letting many chips share a low-latency memory pool to generate output tokens. Jane Street's own deployment, a rack running in its datacenter, is the concrete example given of how a customer puts the system to work. No pricing or licensing terms are stated.
How solid is it
The figures come directly from Etched and are reported by TechCrunch with on-record comment from co-founder and COO Robert Wachen, plus a direct quote from Jane Street's own blog post announcing the round. The valuation numbers and round size are specific and attributed. The performance claims, faster processing and lower costs, are Etched's own characterization of its technology, not independently benchmarked in the source. The exact size of Jane Street's stake in the round is not disclosed, only that Jane Street led it.
Risks and caveats
The source gives no total funding raised to date, no employee count or revenue figures, and no customer beyond Jane Street. It does not say what the new capital will be used for, nor whether Etched has any plans for an IPO or further raises. Rapid valuation step-ups over a single month are inherently a paper figure until backed by revenue or a liquidity event, and the source does not address Etched's revenue or contracted business. Etched's own history, originally positioned as etching one model per chip before pivoting to running any frontier model, is a reminder that the company's technical claims have shifted before.
“We tested the chip and are pleased with the early results. Etched's unique approach to inference delivers the precision we will need to support our most demanding workloads. We're excited to now have our own rack running in our datacenter.”
— Jane Street, in a blog post announcing the round