The startup betting it can out-design Nvidia on inference is suddenly one of the AI chip world's hottest names. This week, The Wall Street Journal reported that Etched is in talks for a valuation of about $20 billion — a striking figure for a company whose chips are only beginning to reach customers.
Two rounds at once
Etched is negotiating a back-to-back financing: one round led by existing investor Jane Street at roughly $20 billion, and a separate round led by Sequoia Capital at around $10 billion. Crucially, neither has closed, and the WSJ cautioned that terms could still change — so this is a target and a negotiation, not a done deal.
A vertical leap
The numbers are remarkable for their velocity. A $20 billion valuation would quadruple the roughly $5 billion mark Etched held after a $500 million round that closed in December 2025. The company has raised about $800 million in total and only exited stealth publicly on June 30, 2026 — selling a stake at one price and almost immediately raising more at a far higher one, a pattern the WSJ calls a signature of this AI cycle.
The bet
Founded in 2022 by Harvard dropouts Gavin Uberti, Chris Zhu and Robert Wachen, Etched builds transformer-ASIC inference chips — silicon hard-wired to the transformer architecture behind today's large models, rather than the general-purpose GPUs Nvidia sells. The wager is that specialization wins on inference economics. Etched says it has around $1 billion in customer demand lined up, though large-scale commercial deliveries have not yet begun.
The risk
That is also the risk. A transformer-specific chip is a leveraged bet that the architecture stays dominant; a shift in model design could strand it. Investors are pricing Etched as if inference-specialized silicon is the future — before the company has shipped at scale — which is exactly what makes a $20 billion talk both a vote of confidence and a wager on unproven volume.
