What does Etched build, exactly?

Etched designs and ships complete hardware systems — chips, memory and interconnects — built only to run already-trained AI models faster and cheaper, rather than to train them.

How much has Etched raised, and at what valuation?

Etched raised $700 million at a $21 billion valuation on August 18, 2026 — up from a $10.3 billion valuation in a $300 million Series C just a month earlier, and from $5 billion in December.

Who is using Etched's hardware today?

Jane Street is Etched’s first customer and has installed the company’s first shipped rack. Etched also says it has signed over $1 billion in contracts with other AI companies and cloud providers, though it hasn’t named them.

How is Etched's chip different from a Nvidia GPU?

Etched’s systems are built only for inference, splitting the prompt-reading (“prefill”) and answer-generating (“decode”) stages across specialized hardware linked by a shared memory pool, rather than using one general-purpose chip for both training and inference.

AI Chip Startup Etched Raises $700M at $21B Valuation

Etched, a startup building chips dedicated to running already-trained AI models rather than training them, has raised $700 million at a $21 billion valuation, the company said August 18. The financing more than doubles Etched’s $10.3 billion valuation from a Series C round announced in July, and puts the two-year-old company at roughly quadruple its December valuation.

Jane Street, the quantitative trading firm, led the round after testing Etched’s hardware and buying it outright. “We tested the chip and are pleased with the early results,” the firm said, adding that Etched’s approach “delivers the precision we will need to support our most demanding workloads.” Jane Street has installed Etched’s first shipped rack and is now its first paying customer, alongside more than $1 billion in contracts Etched says it has separately signed with other AI companies and cloud providers.

What Etched actually sells

Unlike Nvidia, whose GPUs handle both training and inference, Etched builds systems purpose-built only for inference — the computing that happens after a model is trained, when it responds to a user’s prompt. Its architecture separates the “prefill” stage, where a chip reads and processes a prompt, from the “decode” stage, where it generates output tokens, and links many chips into a single shared memory pool to move data between them faster. As we’ve explained before, that specialization is Etched’s bet against general-purpose GPUs as inference, rather than training, becomes a growing share of AI compute spending.

The new round, backed by returning investors including Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global and Blackstone, comes about a month after Etched was reported to be in talks for a $20 billion valuation — a figure investors have since blown past. It’s the latest sign of investor appetite for AI hardware startups betting they can carve out a niche next to Nvidia’s dominance, even as Etched still has to prove it can manufacture and ship chips at scale beyond its first rack.

What Etched actually sells

Frequently asked questions