Etched, the fast-growing AI chip startup behind the transformer-specific Sohu ASIC, has locked down roughly 80,000 square feet in Milpitas, taking over a one-story industrial building to expand its lab and hardware operations. The deal is the latest sign that chip and AI infrastructure firms are hunting for ground-floor space that can handle dense, power-hungry racks, and it is already forcing local landlords and planners to rethink what counts as move-in-ready industrial real estate.
According to the Silicon Valley Business Journal, Etched’s Milpitas lease covers about 80,000 square feet in a one-story industrial structure and is aimed squarely at scaling testing and deployment. Brokers told the outlet they are tracking companies that together are seeking roughly 12,000,000 square feet of similarly power-capable space across the region, a wish list that would remake the industrial map of Silicon Valley if even a slice of it lands.
Etched Is Scaling Hardware Fast
Etched has been building out hardware and raising substantial venture funding as it moves from prototyping toward larger deployments. In an interview with TechCrunch, CEO Gavin Uberti said the team “made a bet that transformers would take over the world,” a strategy that helps explain why the startup now needs far more floor area and electrical capacity than a typical software outfit. The company’s Sohu ASIC promises much higher inference throughput per rack, which makes dense, rack-level deployments practical for customers and for Etched’s own validation work.
Why Power And Amps Matter
Many older Bay Area industrial buildings were wired for more modest loads, about 2,000 amps per building in some cases, which is often insufficient for racks packed with inference servers, as reported by the Silicon Valley Business Journal. Upgrading service typically requires new switchgear, transformers or even utility line extensions, and can add weeks or months to a move-in timeline while piling on costs and permitting hurdles, according to commercial power contractors. Optimum Energy Services and other engineering firms warn that utility coordination and transformer work are the usual bottlenecks when converting older warehouses for compute-dense uses.
Market Pressure In Silicon Valley…