AI has fused compute strategy with energy strategy. Power access determines where AI can scale.
Global Data Center Power Demand
- 2024 Baseline: 415 TWh (1.5% global electricity)
- 2030 Projected: 945 TWh (+128%)
The 7-Year Infrastructure Gap
- Grid Connection Wait: 7 Years
- Nuclear Plant Lead Time: 10+ Years
AI demand growing in months, grid capacity in years.
Utility Investment Response
- 2025-2029 CapEx: $1 TRILLION
- US Grid Annual: $50B/year
Big Tech: Going Direct to Power
Microsoft — Three Mile Island
Restarting nuclear plant, 20-year power purchase agreement
Google — Kairos SMR Deal
500 MW by 2030, first commercial SMR, $4.75B Intersect Power acquisition
Amazon — Nuclear SMR Portfolio
Multiple nuclear investments, data center nuclear as part of $125B infrastructure spend
Meta — Nuclear RFP
Seeking 1-4 GW nuclear power deal, carbon-free baseload for AI
Why Big Tech Is Going Nuclear
- 24/7 Baseload: AI training needs constant power, not intermittent
- Carbon-Free: Meets sustainability commitments at scale
- Land Efficient: Nuclear: 1 acre/MW vs Solar: 5-10 acres/MW
Sovereign AI Power Projects
- UAE AI Campus: 5 GW
- Saudi Humain Project: Multi-GW
- Stargate (US): 6 GW+
GPU Power Escalation
H100: 700W → B200: 1000W → GB200: 2.7kW → NVL72: 120kW
Energy = New Competitive Moat. Those with power access can deploy unlimited AI capacity. Those without are stuck in 7-year grid queues.
This is part of a comprehensive analysis. Read the full analysis on The Business Engineer.
More Resources
