Doubling Down Twice: Etched Fields Funding Bids at $40–50 Billion, Just Weeks After Its $21 Billion Round

Barely six weeks after closing a $700 million round at a $21 billion valuation, AI chip startup Etched is already fielding new investment offers that would double its price tag again — or more. According to a TechCrunch report published October 5, people familiar with the company say Etched is reviewing incoming bids ranging from $40 billion from top-tier investors to $50 billion from lesser-known backers. The talks are early, terms could change, and Etched declined to comment. But the signal is unmistakable: the market’s appetite for anything that promises to break Nvidia’s stranglehold on AI inference has not cooled — it has accelerated.

From Harvard Math Class to $40 Billion in Four Years

Etched’s story has become startup lore by now. Co-founders Gavin Uberti and Chris Zhu met in an advanced math course at Harvard; Robert Wachen, now COO, was Uberti’s roommate. All dropped out to chase a single contrarian thesis: instead of building another general-purpose GPU, design silicon that runs one thing — transformer models — and nothing else.

That bet looks considerably less contrarian in 2026. The company’s valuation history reads like a stairway:

• July 2026: $300 million round at a $10.3 billion valuation, led by Sequoia
• September 2026: $700 million round at a $21 billion valuation, with quant trading firm Jane Street leading
• October 2026: bids now on the table at $40–50 billion, per TechCrunch’s sources

Back-to-back rounds at leaping valuations — essentially one financing split into tranches — have become a hallmark of the buzziest AI startups. Etched has now done it twice, and may be about to do it a third time within a single calendar year. A person familiar with the offers told TechCrunch that if Etched raises as much as its last round, it could bank as much as 3.5 years of runway.

Why Investors Are Hot to Own a Piece

The enthusiasm rests on four concrete pillars rather than pure narrative.

Real customer orders. In July, Etched said it had secured $1 billion in customer orders, including a system delivered to Jane Street — which is both an investor and a customer, an unusually tight alignment. For a quant firm, a microscopic speed edge in inference translates directly into trading profits, which explains why Jane Street put capital in and took delivery of early hardware.

Manufacturing proof. Etched manufactured its test chip at a TSMC factory this summer — a critical de-risking milestone that separates it from chip startups that never leave simulation. The company also built a 10-megawatt data center in Silicon Valley and established a Taiwan facility to coordinate production near TSMC. It sells full rack-scale inference systems, not just silicon.

Talent draw. Roughly 15% of Etched’s 400-person workforce previously worked at Nvidia, according to The Wall Street Journal. In a talent war where the incumbent’s own engineers are the scarcest asset, that ratio is a quiet but powerful endorsement.

A cost-and-speed claim that matters. Etched designed two new components from scratch to accelerate inference — the computation that happens after a user submits a prompt — and claims its chips process more tokens faster and cheaper than Nvidia’s. If that holds at rack scale, it attacks the single largest line item in every AI company’s budget.

The Wider Pattern: Inference Is the New Battlefield

The Etched bidding war doesn’t exist in isolation. The same week, AMD closed its $8.2 billion all-stock acquisition of Fei-Fei Li’s World Labs — its second-biggest deal ever — and CME began trading GPU futures. Capital is rotating from “who can build the biggest model” to “who can serve tokens at the lowest marginal cost.” Inference is where the recurring revenue lives, and purpose-built silicon is the most direct weapon against Nvidia’s roughly dominant share of that market.

Etched is not alone in the lane — Groq, Cerebras, and the rack-scale efforts of the hyperscalers all chase the same prize. But few have moved from test chip to $1 billion in orders to $40-plus billion bids this fast.

What Could Go Wrong

The risks are as real as the momentum. Talks are early and may not close at all; “lesser-known backers” offering the highest number is often a sign that quality investors are anchoring lower. Serving transformer architectures means Etched’s advantage is architecturally specific — a decisive shift in model design could erode its edge. And competing with Nvidia means competing with a company that controls the software ecosystem (CUDA), the supply relationships, and the balance sheet.

Still, the direction of travel is clear. Six weeks ago Etched was worth $21 billion. Today, people close to the company say the number on the table starts at $40 billion. Whether this round closes at that price or not, the inference-silicon land grab has entered its most expensive phase — and Etched is, for now, the most-watched claim in it.