Etched just pulled off one of the fastest valuation jumps in AI hardware history. The startup raised $700 million at a $21 billion valuation, led by Jane Street — the notoriously selective quant fund that actually tested the chips before writing the check. That’s the detail that separates this round from the usual hype cycle. Jane Street doesn’t throw money at narratives.
The numbers tell a wild story. Etched was valued at $5 billion in December. Then $10.3 billion in July after a $300 million Series C. Now $21 billion barely a month later. Investors doubled the valuation in roughly 30 days. That’s not normal growth — that’s conviction pricing based on silicon that exists and performs, not a slide deck.
Co-founder and COO Robert Wachen explained the technical bet to TechCrunch. Inference splits into two stages: prefill, where the system parses your prompt and context, and decode, where it generates the actual tokens you see. Etched attacked both. The prefill chip runs at low voltage, which lets the company pack in more transistors without melting the thing — more tokens processed, faster. For decode, they built what they call cluster-scale memory: a shared, low-latency memory pool that lets many chips work together. The pitch is higher throughput at lower cost than Nvidia’s AI factories.
Etched still fights an old perception problem — the name suggests chips etched for one specific model, which was the original plan but hasn’t been true for a while. Their systems now run any frontier model. Jane Street’s statement in the announcement blog was unusually direct: “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.” The fund already has its own rack running in its datacenter. With Kleiner Perkins, Sequoia, a16z, and Blackstone also on the cap table, the inference hardware race just got a lot more interesting — and Nvidia’s dominance in that layer is no longer a foregone conclusion.
