Etched, the startup building chips dedicated to running AI inference rather than training, is in talks to raise new funding at a valuation of roughly $20 billion. A separate round led by Sequoia Capital values the company at around $10 billion. The two figures point to a financing structure with more than one tranche, and together they mark an extraordinary jump for a company that closed its last disclosed round at a $5 billion valuation only seven months earlier.
From near-collapse to a two-tier raise
Etched was founded in 2022 by Harvard dropouts and Thiel Fellows Gavin Uberti and Robert Wachen. By 2023 the company was nearly out of cash, unable to get investors on the phone in an AI chip market still fixated on training rather than inference. That changed once its transformer-only inference chip, Sohu, worked on TSMC’s manufacturing line and the company began booking customer orders.
Etched has now raised a combined $800 million across a string of previously undisclosed rounds, including a $500 million tranche that closed in December 2025 at a $5 billion post-money valuation, led by Stripes. Backers from that round include Peter Thiel, Jane Street, Hudson River Trading, Two Sigma, Stanley Druckenmiller, and VentureTech Alliance, a fund tied to TSMC. Angel investors include Andrej Karpathy, Geoffrey Hinton, Fei-Fei Li, and Arthur Mensch.
Why two valuations for one financing event
Reporting two different price tags on the same round is not unusual in the current venture market. A lead investor sometimes puts a larger share of capital in at a lower, preferential valuation while adding a smaller amount at a much higher headline number, which produces a public figure well above the round’s actual blended price. That mechanism has drawn public criticism from founders at other AI startups in recent months, and it fits the pattern here: a $10 billion Sequoia-led tranche alongside talk of a $20 billion valuation for the wider raise.
The commercial case behind the number
Unlike many chip startups still selling a roadmap, Etched says it has already booked more than $1 billion in signed customer contracts for what it calls frontier inference clusters, bundles of its Sohu chips with custom racks and software built around them. The company says its first chip worked on the first manufacturing pass at TSMC, a rare outcome in chip design, and that early systems are already running models including DeepSeek, Qwen, Mamba, and Llama for test customers ahead of shipments this summer.
That combination, a working chip, a foundry relationship with TSMC, and real contract value, is what investors are pricing at four times the company’s valuation from seven months ago. It also reflects a broader shift in AI infrastructure spending, as more capital moves toward inference-specific silicon built to challenge Nvidia’s position once models are deployed rather than trained.
Leave a Reply