The offers
Etched is reviewing funding bids that value it between $40 billion and $50 billion, TechCrunch reported on 5 October, citing a person familiar with the company. The lower end comes from top-tier investors and the higher end from lesser-known backers. The talks are early and the terms could change. Etched declined to comment.
The sequence behind that number is the story. In July the company raised $300 million at a $10.3 billion valuation, led by Sequoia. A $700 million round at $21 billion, led by the quantitative trading firm Jane Street, followed shortly after. A round at $40 billion would be roughly double that again.
What the company actually sells
Etched builds AI hardware systems around its own chips, aimed at inference — the processing that happens after a user submits a prompt, rather than during training. Co-founder and chief operating officer Robert Wachen said the company designed two new components from scratch to speed that up. Etched claims its chips handle more tokens, faster and more cheaply, than Nvidia’s. That is the company’s own claim and has not been independently benchmarked.
In July it said it had $1 billion in customer orders, Jane Street among them, after manufacturing its test chip at a TSMC plant. It runs a 10-megawatt data centre in Silicon Valley and a facility in Taiwan to coordinate production close to TSMC.
The shape of the company
Etched is about four years old and has roughly 400 employees. The Wall Street Journal has reported that about 15% of them previously worked at Nvidia, which is both a hiring strategy and a description of where the expertise sits.
The founders are Gavin Uberti and Chris Zhu, who met in an advanced mathematics course at Harvard, and Robert Wachen, who was Uberti’s roommate. All three dropped out to start the company.
What the money is for
One source told TechCrunch that raising as much as the last round could give Etched up to three and a half years of runway. That is the number worth holding onto, because a custom-silicon company’s problem is time, not valuation: a chip design has to survive a tape-out, a manufacturing slot, a bring-up and a customer deployment before anyone finds out whether the architecture was right.
Three and a half years is roughly two of those cycles. It is also long enough to outlast whatever Nvidia ships next, which is the bet an inference-specific chip has to make.
What to watch
Whether a round actually closes and at what number, whether Etched publishes inference benchmarks that someone outside the company can reproduce, and whether the $1 billion order book converts into revenue. A backlog signed before a chip ships is a statement of intent, not a sale.
