Etched Bücher $ 1B in Bestellungen für benutzerdefinierte Inferenzcluster, Treffer $ 5B Bewertung

Etched has booked $1 billion in contract orders for its custom inference systems before even finishing customer testing—a signal that the AI industry is desperate enough for cheaper, faster inference to bet on unproven hardware.

Why Inference Chips Command a $5B Valuation

TSMC manufactured Etched's first chip earlier this year. The startup is now testing what it calls “frontier inference clusters”—bundles of its custom silicon, custom racks, and software designed to run inference faster and cheaper than any general-purpose GPU. Inference is the single biggest cost center and bottleneck for AI companies serving users at scale. Every hyperscaler (Amazon, Google, Microsoft) and even OpenAI itself is building or buying custom chips to attack that problem. Etched’s $1B order book says the market is willing to pay for a solution today, not wait for Nvidia’s roadmap.

Etched’s Road from Near-Death to $500M Round

Co-founders Gavin Uberti and Robert Wachen dropped out of Harvard and became Thiel fellows to start Etched in 2022. By 2023, they couldn’t get a meeting: every major investor passed on a 30-page memo arguing that AI would need specialized chips. The company was operating month-to-month, nearly out of cash. Fast-forward: Etched has now raised $800 million total, most recently a $500 million round at a $5 billion post-money valuation closed in December 2025. The investor list reads like a who’s-who of quantitative finance and AI: VentureTech Alliance, Jane Street, Hudson River Trading, Two Sigma, Ribbit Capital, plus angel bets from Andrej Karpathy, Geoffrey Hinton, Fei-Fei Li, Arthur Mensch, Scott Wu, Stanley Druckenmiller, and Peter Thiel.

What $1B in Orders Says About the AI Infrastructure Shift

Those orders aren’t for chips alone—they’re for complete systems. That tells me customers want a drop-in replacement for GPU clusters, not just a component to integrate. Competitors are already moving: Cerebras had the first breakout AI IPO of the year, Groq raised $650 million, and hyperscalers are building in-house. Etched’s edge is that it bet early on a pure-inference architecture when everyone else was chasing training. If their clusters deliver on claimed power efficiency and performance, the next battleground in AI won’t be training—it’ll be serving models at scale, and Etched might own the lane.

Source: Nvidia competitor Etched hits $5B valuation, $1B in sales for AI chip
Domain: techcrunch.com