SemiAnalysis: China has 24 GW of operational compute capacity and another 50 GW planned or under construction, compared to 56 GW operational in the US
THE SO WHAT
China planning 50 GW on top of 24 GW operational compute versus 56 GW in the US reframes AI as a power race, not just a chip race. For global operators, data residency and model deployment decisions will increasingly track where cheap, reliable gigawatts—not just GPUs—are available.
READ THE SOURCE
MORE FROM THE WIRE
Deep & Emerging TechFinland’s Data Center Tally Tops €67 Billion on AI Boom
€67B of data center commitments in Finland is energy arbitrage at scale—cold climate, stable grid, and political predictability are now core AI assets. If you're planning large deployments, treat geography and power contracts as first-class variables alongside GPU supply.
Deep & Emerging TechA look at Emeryville, CA-based Atomic Machines, which is training AI on materials and designs to revamp how microelectromechanical systems (MEMS) are built
AI-driven design for MEMS and micro-robots collapses iteration cycles in a part of the stack that’s been slow and artisanal—this is software eating fabrication recipes. Hardware teams should start mapping where generative design and simulation can replace manual CAD and trial-and-error in their own component pipelines.
Deep & Emerging TechTSMC’s Quarterly Revenue Jumps 51% After AI Demand Holds Up
A 51% revenue jump at TSMC tied to AI demand says the infra buildout is still in full swing—capacity, not appetite, is the binding constraint. If you depend on advanced nodes, lock in multi-year supply and diversify designs now rather than betting on prices normalizing quickly.
Deep & Emerging TechItaly Data Center Group Sees Investments Reaching €37 Billion
€36.9B of projected Italian data center spend through 2036 means AI demand is now reshaping national power, land, and permitting markets. If you rely on European compute, start mapping latency and regulatory tradeoffs across Italy versus traditional hubs like Frankfurt and Amsterdam — the cheapest capacity may not be where your stack lives today.