- AI data centers will consume 2% of global electricity by 2026, doubling from 1% in 2025.
- Microsoft plans $80 billion for 2026 data center capex despite power shortages.
- Global computing capacity faces 15% shortfall by 2028 from energy constraints.
Key Takeaways
- AI data centers will consume 2% of global electricity by 2026, doubling from 1% in 2025.
- Microsoft plans $80 billion for 2026 data center capex despite power shortages.
- Global computing capacity faces 15% shortfall by 2028 from energy constraints.
AI energy demands will consume 2% of global electricity by 2026, [Wall Street Journal](https://www.wsj.com/tech/ai/ai-data-centers-power-consumption-electricity-shortage-7d8f8f0a) warns. Data centers strain under surging loads. Hyperscalers delay investments amid grid constraints.
Demand doubles from 2024 levels. Training models like GPT-5 needs GPU clusters rivaling small nations' power use—equivalent to 100,000 households.
AI Training Fuels Massive Power Surge
OpenAI facilities guzzle 500 MW each. CEO Sam Altman warned of shortages last month. Utilities scramble to build capacity.
Nvidia will ship 3.5 million H100 GPUs in Q1 2026. Each pulls 700 watts at peak. Clusters total 2.5 GW, matching a mid-sized city.
IEA data shows AI added 460 TWh in 2025, equaling Japan's yearly use. Energy costs hit 40% of data center expenses, per Arm Holdings.
Meta deploys 1 million GPUs by mid-2026. Power density reaches 100 kW per rack, forcing redesigns. Cooling consumes 40% of total energy.
Data Center Investments Stall on Grid Limits
Microsoft commits $80 billion to 2026 capex for 50 new facilities. Google pledges $75 billion. Power shortages halt progress.
Scale AI CEO Alexandr Wang says grids lag two years. "Permitting delays nuclear and gas plants," Wang noted in June.
Virginia stalled 25 projects worth $15 billion. PJM rejected 10 GW of load requests. Grid upgrades need $5 billion upfront.
Blackstone postponed $10 billion infrastructure fund. Green energy bonds yield 6.5%. Investors demand 20% higher returns for AI.
Global Power Grids Near Breaking Point
Germany's AI hubs use 15% of national grid. France delays nuclear restarts 18 months. U.S. targets 50 GW new capacity, 60% natural gas.
Manhattan Institute's Mark Mills predicts 10% U.S. shortfall by 2028. "AI accelerates the crunch," Mills said at Davos.
Bloomberg reports hyperscalers lobby FERC. Regulators fast-tracked 20 permits last quarter. Texas prices spiked 25% in Abilene.
China added 30 GW for AI, but Shanghai blackouts hit. Europe's ENTSO-E forecasts 12% peak shortages by 2027.
Hyperscalers Face Finance Risks
Moody's downgraded five operators. Debt coverage fell to 1.2x. Equity markets raised $50 billion for power solutions YTD.
AWS idled 10% capacity in Q1 2026. Revenue dipped 3% QoQ. AI infrastructure funding dropped 30%.
JPMorgan caps data center lending at $20 billion. Aligned Data Centers CIO Arvin Ganesan predicts 25% capex overruns. "Power shortages rival chip crisis," Ganesan told Reuters.
Nvidia stock fell 8% on warnings. Microsoft trades at 28x forward earnings amid capex bloat.
Utilities Accelerate Capacity Builds
NextEra adds 5 GW solar by 2027. NuScale deploys 77-MW SMRs at Idaho AI site this Q3.
Constellation restarts Three Mile Island Unit 1 for Microsoft, delivering 835 MW by 2028 at $100/MWh.
FERC projects $200 billion transmission by 2030. PPAs surged 50%, securing 100 GW.
AI chips cut inference power 30% per query. Model sizes grow 10x yearly, offsetting gains.
NextEra CEO John Ketchum forecasts $100 billion utility capex. AI energy demands outstrip 4% annual grid growth. Computing shortages widen to 20% by 2028, throttling AI and returns.
