Why the Power Grid, Not the Chip Supply, Is Now AI's Biggest Constraint
Key Takeaways
- Severe heatwaves and rapid AI data centre expansion are colliding over limited public power grid capacity.
- Megawatts and interconnection availability have superseded chip supply as the dominant constraint for AI scale.
- Long 10-to-15-year leases cushion data centre real estate, but utility interconnection delays create new friction.
- Behind-the-meter power solutions and on-site generation are transitioning from luxury to essential operational requirements.
Two power stories have been running in parallel through 2026, and this week they collided.
The Human Side of the Strain
Intense heat is straining electrical grids across North America, putting roughly 100 million people at risk of capacity strain, rolling brownouts, and rising bills.
The Industrial Side of the Strain
At the same time, the AI infrastructure buildout — data centres, chip fabrication, cooling systems — is competing for that same finite grid capacity. Data centres are now, quite literally, competing with air conditioners.
Why Real Estate Fundamentals May Still Hold
One data centre REIT CEO argued this week that an AI capex slowdown is "not the end of the world" for real estate, pointing to 10-to-15-year tenant leases that outlast any single news cycle.
The New Planning Problem for Utilities
Utilities must now plan for peak human demand and peak industrial demand converging at once — a materially harder problem than planning for either alone.
What This Means for Investors
Behind-the-meter power — on-site generation, battery storage, direct power purchase agreements — is shifting from a competitive advantage to a baseline resilience requirement.
