Microsoft CEO: Not enough electricity for all AI GPUs in inventory (www.tomshardware.com)

Microsoft CEO Satya Nadella told a podcast that the AI industry’s current bottleneck isn’t a surplus of GPUs but insufficient power and data‑center “warm shells” to actually deploy them. He said Microsoft has AI accelerators sitting in inventory that it can’t plug in because buildings lack the necessary power, cooling and infrastructure. That reframes the compute shortage narrative: even as Nvidia ramped GPU supply, scaling AI now depends on large, reliable electricity and facility buildouts — a constraint that’s prompting calls for massive new generation (OpenAI wants 100 GW/year) and interest in small modular reactors and other big energy projects. The technical and market implications are significant. “Shells” are empty data‑center buildings provisioned with power and water; without them, raw GPUs and racks can’t be energized. The power demand also raises policy and cost issues (higher consumer bills, geopolitical implications where China’s hydro and nuclear capacity give it an edge). At the same time, OpenAI’s Sam Altman noted a future where highly distilled, quantized models could run locally on low‑power consumer devices (e.g., aggressive 4‑bit models), which would lessen inference demand on large clusters — though high‑precision training and premium inference would likely keep centralized data centers relevant. Together these trends complicate billion‑dollar datacenter bets and reshape how the AI economy must plan for energy, cooling and supply chain realities.