Google just signed a 22-year contract to buy up to half the output of a Finnish nuclear plant. Not for the reactor — for the electrons.
This month Alphabet committed at least €13 billion ($15.1B) to three new data centers in Finland, including a power-purchase agreement covering 50% of Fortum's Loviisa plant. Ruth Porat, Google's president and chief investment officer, calls it "BYOP — bring your own power." It's Google's first nuclear deal outside the US, but it's hardly the first in the industry. Meta has contracted 2,609 MW across TerraPower, Oklo, and Vistra for its "Prometheus" supercluster in Ohio. Oklo and NVIDIA are now co-developing what they call "nuclear-powered AI factories."
The Constraint Is Shifting
Token prices keep dropping — reasoning output that cost $60 per million tokens a year ago now has open-weight rivals under $3. That efficiency is real, but it's being consumed by volume. The bottleneck is moving from silicon to electrons, and energy doesn't obey Moore's Law. A gigawatt is a gigawatt; interconnection queues are measured in years, not quarters. So hyperscalers are vertically integrating into baseload power.
What This Means for Enterprise Buyers
Three consequences follow.
- AI unit economics now have a physical floor. Power cost is becoming the denominator, not token price.
- Site selection is a strategic decision. Finland's edge isn't just cold air — it's stable, low-carbon baseload and a grid that can absorb new load. Your provider's data-center geography is now part of your AI roadmap.
- Cloud contracts are becoming energy contracts. Expect power-backed SLAs, location transparency, and power-tiered pricing.
Stop budgeting AI in tokens alone. Ask providers for power commitments, and treat energy as a first-class input in your AI operating model. The enterprises that secure electrons will outlast the ones that only bought GPUs.