Etched, the AI chip startup building processors specialized for transformer-model inference rather than general-purpose GPU workloads, closed a $300 million Series C at a $10.3 billion valuation -- roughly double the approximately $5 billion mark it held just a few months ago. Jane Street, an existing investor, is separately reported to be in talks to lead an even larger financing at a $20 billion valuation, though that round had not closed as of mid-July.
The jump reflects both Etched's own traction and a broader market repricing of AI chip alternatives. The company says it has booked roughly $1 billion in customer contracts and has around $800 million in total disclosed funding to date, with its pitch centered on hyperscale cloud providers and dedicated AI platform companies that want inference-optimized silicon rather than paying Nvidia's premium for flexible-but-generic GPU capacity.
“Watch whether the reported $20 billion Jane Street round actually closes, and whether Etched names its first major cloud customer publicly.”
Etched competes in an increasingly crowded but still Nvidia-dominated field alongside Cerebras, Groq and a wave of newer entrants, all chasing the same thesis: that as inference volume dwarfs training volume industry-wide, purpose-built chips can undercut general-purpose GPUs on cost per token even if they're less flexible. Roughly $10.7 billion has flowed into seed-through-pre-IPO semiconductor rounds industry-wide in 2026, a sign investors increasingly believe at least some of that thesis will pay off.
The timing compounds the story. Nvidia's own China revenue has fallen to essentially zero amid export restrictions, AMD just won a marquee Microsoft Azure deployment for its Helios rack-scale platform, and Jensen Huang has spent the week publicly defending competitors' access to Chinese AI models rather than talking up Nvidia's own roadmap. For the first time in years, serious late-stage capital is flowing to Nvidia alternatives at a scale that could actually matter to hyperscaler purchasing decisions rather than remaining a venture-stage curiosity.
For VCs, the read is that inference-specific silicon has crossed from thesis to allocatable category -- a $10.3 billion mark for a company that was worth half that a few months ago is the kind of re-rating that pulls more capital into the category fast. The risk is concentration: a huge share of Etched's implied value depends on landing and keeping a handful of hyperscaler-scale customers, and any stumble on manufacturing yield or a competitive chip from Nvidia itself could compress the multiple quickly. Watch whether the reported $20 billion Jane Street round actually closes, and whether Etched names its first major cloud customer publicly.
