The numbers

Etched, which develops inference hardware, announced it has raised another $700 million at a $21 billion valuation. The round was led by Jane Street, the famed quant fund, after it tested and bought the startup's AI hardware.

As TechCrunch notes, even by AI standards this valuation step-up is jaw-droppingly fast:

  • In December the company was valued at $5 billion.
  • In July it raised a $300 million Series C at a $10.3 billion valuation.
  • In August the valuation reached $21 billion — an increase of nearly $11 billion in a month.

What the product is

Etched delivers its technology as full systems it calls "frontier inference clusters". For comparison: its competitor Nvidia calls its own full systems AI factories.

According to co-founder and COO Robert Wachen, investors are so enthusiastic because the company designed two new components from scratch to speed up inference — the computing process that happens after a user submits a prompt.

The two stages of inference

In Wachen's account, inference is built in two stages: prefill and decode.

In the mathematically and compute-intensive prefill phase, the system must understand the prompt, including context. In the memory-intensive decode phase, it generates output tokens, meaning the actual answer the user sees.

The components the company designed map onto that split. For prefill it created a chip that operates at low voltage, allowing it to pack in more transistors without the typical heat problems of other high-end AI chips, and therefore process more tokens faster. For the decode side it created a new type of memory and an interconnect.

How to read it

What makes this notable is who led the round. Jane Street is not a technology investor but a quantitative trading fund; it tested and bought the hardware for its own use first, then invested in the company. Behind the valuation, in other words, sits a customer experience rather than a market forecast. Investor and customer being the same party makes the claim somewhat more concrete than in other rounds.

The speed itself, though, is a warning sign. A valuation quadrupling from December to August may reflect capital piling into inference hardware more than the product maturing. There is a cautionary example in the same space: Groq, which tried to rival Nvidia with its own chip, turned into a cloud company operating Nvidia systems after losing its team, and its valuation halved.

Etched's difference is that a customer who tested and bought the product led the round. That makes the picture sturdier, but it is not sufficient on its own; the real test will be how the frontier inference clusters stand against Nvidia systems in independent benchmarks.