Countries are struggling to meet the rising energy demands of data centers (restofworld.org)

Microsoft’s new hyperscale data center in Colón, Mexico, has been forced to rely on gas-powered generators because it couldn’t connect to the national grid — a vivid example of a wider global problem: data center growth driven by AI and cloud computing is outpacing grid capacity. Mexico faces a projected 48,000 MWh shortfall by 2030, and Microsoft told regulators its generators would supply 70% of the facility’s power for 12 hours per day between Feb–Jul 2025, producing CO2 equivalent to roughly 54,000 households. The International Energy Agency estimates nearly 60% of data center electricity now comes from fossil fuels, with renewables supplying just over a quarter — a gap that risks normalizing polluting backup power even as major tech firms pledge carbon neutrality. Technically and strategically this matters: hyperscalers typically require ~100 MW at full scale, Mexico’s data centers may demand about 1.5 GW by 2030, and Querétaro alone already consumes ~200 MW (≈80% of national data‑center demand). The bottleneck is transmission and distribution — not generation — under the monopoly CFE, with investment in T&D down sharply since 2018 and a government plan that prioritizes households. The upshot for the AI/ML community is clear: expanding compute for training and inference will strain grids, complicate sustainability commitments, raise local air‑quality and health risks, and force tradeoffs between rapid deployment and investments in cleaner grid and energy infrastructure.