Big Tech Is Making ‘Empty Promises’ on 500 Gigawatts of Data Centers It Can’t Deliver

Former DOE loan office head Jigar Shah warned that hyperscalers’ public commitments for data-center capacity (commonly cited as ~500 GW) are likely unattainable, arguing physical constraints (grid, transformers, GPUs, memory, CPUs) limit realistic builds to roughly 50 GW by 2030. The piece highlights regional bottlenecks (ERCOT’s ~300 GW load queue vs. a ~70 GW system; Shah sees ~30 GW realistically) and contrasts those limits with massive hyperscaler capex — Microsoft reported Q3 FY26 capex of $30.876B (AI run rate $37B), Amazon spent $44.2B in Q1 FY26 and guided ~ $200B for 2026, Alphabet and Meta also raising capex guidance — and surging NVIDIA data-center revenue ($39.1B, +73% YoY). Market impact: revenue ramps for hyperscalers could be delayed if capacity lags, pressuring shares on heavy capex, while suppliers like NVIDIA stand to benefit from scarcity. Investors should monitor transformer lead times, interconnection queues (e.g., ERCOT), and hyperscaler commentary for signs the announced buildouts are real or merely promotional.