ByteBulletin

[funding] · · 2 min read

Etched raises $700M at $21B valuation as Jane Street bets big on AI inference hardware

The AI chip startup's valuation doubled in a month as its novel inference approach wins over a quant fund with demanding workloads.

By ByteBulletin Editors · Editorial Team

[funding]

Etched has raised another $700 million at a $21 billion valuation, the company announced on Tuesday, in a round led by Jane Street. The famed quantitative trading firm tested the startup's AI hardware—and bought some—before leading the round. The new funding nearly doubles Etched's valuation from the $10.3 billion it commanded in July, and that was already a steep jump from the $5 billion valuation it carried in December.

The breakneck pace reflects investor enthusiasm for Etched's approach to inference—the compute stage that runs after a user submits a prompt. Co-founder and COO Robert Wachen explains that inference happens in two phases: prefill, where the system processes the prompt and its context, and decode, where it generates the output tokens that form the answer. Etched has designed custom silicon and memory for each phase: a low-voltage prefill chip that packs in more transistors without the usual heat issues, and a new type of memory and interconnect for decode that Wachen calls "cluster-scale memory." This setup lets many chips share a memory pool with very low latency, promising higher speeds and lower costs.

A key part of the pitch is that Etched's full systems—which it calls "frontier inference clusters" (competitor Nvidia calls its equivalents "AI factories")—can run any frontier model. That's a shift from the company's early reputation, when it wanted to hard-wire chips to specific models. Wachen says that's no longer the case, though perception is slow to change.

Jane Street's enthusiasm is notable because the firm is known for demanding performance and precision. In a blog post, Jane Street said: "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." That's a strong endorsement from a company that doesn't spend money on hardware lightly.

Other investors in the round include Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone.

SHARE

← All stories