Etched’s $700 Million Round Puts a $21 Billion Price on Inference Hardware

Etched has raised $700 million at a $21 billion valuation, doubling the price investors put on the AI hardware startup a month after its $300 million Series C. Jane Street led the new round after testing and buying Etched hardware, tying the financing to an early customer deployment as the company prepares to fund production.

The move is unusually fast even for a venture-backed AI company. Etched was valued at $5 billion in December, then raised its July Series C at a $10.3 billion valuation. The latest financing puts its valuation up nearly $11 billion from that July mark.

Jane Street’s role goes beyond writing a check. Etched says the quantitative trading firm tested its hardware and has a rack running in its data center; Jane Street said it was pleased with the early results. That makes the firm both lead investor and a publicly identified customer in this financing announcement.

The deployment gives Etched a customer endorsement, but it is not a public performance benchmark. Jane Street described its results as early, and Etched’s claims about faster or cheaper model operation remain company performance claims rather than independently measured results.

Etched delivers what it calls frontier inference clusters: full systems for inference, the computation that runs after a user submits a prompt. The company’s pitch is that this specialization can make models faster and cheaper to run than more general-purpose AI hardware.

The company says its systems can run any frontier model, rather than being custom-designed for one model. That is a shift from Etched’s original intention and addresses a perception that its chips were tied to a single frontier model.

Its technical case rests on two stages of inference

• Prefill: the compute-heavy work of reading a prompt and its context. Etched says its low-voltage prefill chip can pack in more transistors without the usual heat problems, enabling faster token processing.
• Decode: the memory-heavy work of generating an answer token by token. Etched says its cluster-scale memory lets many chips access a shared memory pool with low latency.

Kleiner Perkins, Sequoia, Andreessen Horowitz and Tiger Global participated in the new round. The prior $300 million Series C was led by Sequoia, with Andreessen Horowitz, SK Hynix, Jane Street and Diffusion Capital also joining that financing.

Etched says the new capital will finance production. It also says it has raised $1.9 billion to date and secured more than $1 billion in customer contracts across public and private AI companies and cloud providers. Those are company-reported commercial figures, while Jane Street is the customer publicly identified in this financing announcement.

The valuation now rests on more than a chip design. Etched needs to turn a tested rack, production capital and its stated customer contracts into systems that can be delivered repeatedly. Whether its specialized architecture produces the promised speed and cost advantages across broader deployments remains the central operating test.