Etched, a startup building chips designed specifically to run AI models faster than general-purpose processors, is reviewing funding offers that would value the company at $40 billion to $50 billion, according to people familiar with the discussions cited by TechCrunch. If the round closes near that range, it would roughly double Etched's valuation again in a matter of weeks.
The four-year-old company has climbed through a string of fundraises this year: a $300 million Series C in July valued it at $10.3 billion, led by Sequoia Capital, and a $700 million round shortly after valued it at $21 billion, led by Jane Street, which is also a customer of Etched's hardware. Other backers in that round included Andreessen Horowitz, Tiger Global, Bain Capital Ventures, Kleiner Perkins, Blackstone, Peter Thiel and SK Hynix, the memory-chip maker that supplies the high-bandwidth memory Etched's systems depend on.
Betting Against General-Purpose GPUs
Etched's pitch centers on Sohu, a chip built specifically to accelerate inference — the computation that happens after a user submits a prompt to an already-trained AI model — rather than the training process GPUs are more commonly used for. The company says Sohu processes tokens faster and at lower cost than Nvidia's chips for that narrower job, a specialization-over-flexibility bet that has drawn skepticism from engineers who note Etched has yet to ship the chip to customers at volume even as its paper valuation soars.
Co-founder and chief operating officer Robert Wachen has described the company's strategy as a commitment to controlling the entire hardware stack rather than just the chip itself.
We're deep believers in vertical integration. It's not enough to build a chip… you have to build a machine that can produce the best clusters in the world at gigawatt scale.
Robert Wachen, co-founder and COO, Etched
Etched says it has already booked roughly $1 billion in customer orders after manufacturing test chips through TSMC, and that it operates a 10-megawatt data center in Silicon Valley alongside a production-coordination operation in Taiwan. Its systems are built to run popular open models including DeepSeek, Qwen, Mamba and Llama. About 15 percent of its roughly 400-person staff previously worked at Nvidia, the company it is positioning itself against.
Part of a Broader Scramble for Inference Capacity
The eye-popping valuation jump reflects a broader rush of capital into specialized AI-infrastructure startups as cloud providers and AI labs look for alternatives to Nvidia's chips, which remain in tight supply and command high prices. Investors are effectively betting that demand for running already-trained AI models cheaply at scale — rather than training ever-larger ones — will be the more durable, high-volume business over the next several years.
The funding round has not closed, and the final size and valuation could still shift as Etched weighs competing offers from both established venture firms and newer entrants to its cap table. Whether the company can convert that paper valuation into shipped hardware at scale remains the central question hanging over the deal.
