An AI chip startup just tripled its valuation in six months. Here's who's paying and why.

Key points

• Etched raised $700 million in a funding round announced Tuesday, valuing the AI chip startup at $21 billion.
• Jane Street, a quantitative trading firm known for using algorithms to make financial bets at enormous scale, led the round after testing Etched's hardware in its own data centre.
• Etched was valued at $10.3 billion in July 2025 and $5 billion in December 2024, meaning the company's valuation has roughly doubled in about one month.
• Etched's chips are built specifically to speed up inference, the computing work an AI model does the moment you submit a prompt.
• Backers include Kleiner Perkins, Sequoia Capital, Tiger Global, Blackstone and Peter Thiel, among others.

Barely a month ago, Etched was worth $10.3 billion after a $300 million Series C in July 2025. It was worth $5 billion the December before that. Now investors have handed it another $700 million at a $21 billion valuation. That's nearly $11 billion added in four weeks, which is fast even by AI standards.

The lead investor is Jane Street, a quantitative trading firm that uses sophisticated computer models to trade financial assets. It's not a typical Silicon Valley venture fund, and it didn't behave like one: Jane Street tested the hardware before committing, then installed a rack of Etched chips in its own data centre.

"We tested the chip and are pleased with the early results," Jane Street wrote in the announcement. "Etched's unique approach to inference delivers the precision we will need to support our most demanding workloads."

We first covered Etched on 23 July 2026, a period when the broader chip story was already moving fast: Anthropic was hiring chip designers and Groq had pivoted to Nvidia hardware rather than compete head-on.

What does Etched's hardware actually do?

Etched builds specialised chips packaged into full server systems it calls frontier inference clusters. Inference is the computing work that happens after you hit send on a prompt: the model reads your question and writes back an answer.

That work splits into two stages. The prefill phase is where the system reads and processes your prompt. The decode phase is where it writes the answer, token by token (a token is roughly a word or part of a word). Etched built a separate custom chip for each. The prefill chip runs at low voltage, so it can pack in more transistors (the tiny switches that do computing work) without overheating, which means it handles more text at once. For decode, Etched built a shared memory system it calls cluster-scale memory, letting many chips pool their resources as if they were one.

Co-founder and COO Robert Wachen told TechCrunch the combination delivers higher processing speeds at lower cost per query.

What's worth watching here?

Jane Street's decision to test before investing matters. A firm of that size running Etched chips in production is a commercial signal, not just a bet on a pitch deck. Survivorship bias is real in AI hardware, though: for every chip startup that lands a marquee customer, several others raise at sky-high valuations and quietly disappear.

Etched is also still shaking off an early reputation for locking chips to a single AI model. It now says its systems run any frontier model, which matters if the field shifts. With software-only inference acceleration and Nvidia's own data centre bets all competing for the same inference budgets, Etched still has to prove its approach wins at scale.

One honest takeaway: The cost of running AI models is a real line item for any business that pays inference bills to providers like OpenAI or Google. More chip options tend to lower prices over time. Watch whether Jane Street's early results hold up in production before reading this valuation as a verdict.