Startup Etched announced that it had raised $700 million in a new round led by Jane Street, after the latter tested the first complete artificial intelligence system shipped by Etched and purchased it to operate in its data center. The round raised the company’s valuation to $21 billion, a notable jump that came just one month after a previous round.
Etched had reached a valuation of $5 billion in December, then raised $300 million in a Series C round in July at a valuation of $10.3 billion. Its valuation therefore increased by about $11 billion in one month, reaching more than twice its previous level.
A Bet on the Inference Phase
Etched presents its technology as integrated systems it calls “frontier inference clusters,” rather than limiting itself to selling individual chips. Its design focuses on the inference process that begins after a user sends a request to an artificial intelligence model—the stage in which the request is understood and the answer is generated.
Robert Wachen, the company’s co-founder and chief operating officer, explained that inference consists of the “prefill” and “decode” stages. Prefill requires a large amount of computation to understand the request and its context, while the decode stage relies more heavily on memory because it produces the tokens that form the user’s final answer.
Two New Components to Accelerate Processing
Etched says it designed two components from scratch to address the requirements of the two stages. The prefill chip operates at low power, allowing a greater number of transistors to be packed in without the thermal problems typical of high-performance AI chips. According to the company, this enables a larger number of tokens to be processed at higher speed.
For the decode stage, the company developed a new type of memory and an interconnect that it calls “cluster-level memory.” This architecture enables multiple chips to communicate and use a shared memory pool at high speed and with low latency, which Etched says could increase speed and reduce cost.
Why Does Jane Street’s Investment Matter?
Jane Street’s role in the deal is not limited to financing; its announcement indicates that it tested the chip and found its initial results satisfactory, then installed a dedicated rack for it in its data center. This step gives the startup a practical test from an entity that uses demanding computational workloads, while the round reflects investors’ confidence that improving inference can be an independent competitive field within AI infrastructure.
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.
Etched is also attempting to correct a perception that arose during its early stages: that the company “etches” a specific model into its chips, with each chip designed to run one advanced model. The company says this is no longer its current approach and that its systems are capable of running any advanced model.