Larry Fink Says AI Needs More Electricity. Google Is Trying To Need Less Of It.

The market is still pricing AI as a silicon/software problem, but the binding constraint is increasingly grid capacity and power procurement. That shifts incremental economics away from model vendors and toward the owners of transmission, switchgear, transformers, gas turbines, and on-site generation; the best-positioned vendors are the ones with multi-year backlog and pricing power, not the ones selling compute on thin margins.

For hyperscalers, electricity scarcity is not a demand killer, but it does raise the cost of growth and can delay capacity monetization. Over the next 1-3 months, the key risk for GOOGL/GOOG is not a narrative miss on AI usage; it is capex intensity outpacing revenue inflection if power-secured capacity becomes the gating item. If management leans on more leases, PPAs, or captive generation, cloud margins can compress before the market sees any slowdown in customer demand.

Contrarian take: this is less bearish for AI adoption than it is bullish for the picks-and-shovels layer that the market has not fully re-rated. The second-order winners are companies that sell the electrical plumbing into data centers, while the biggest losers are hyperscalers forced to compete for scarce megawatts. The thesis breaks if Alphabet shows it has already locked in low-cost power at scale and can keep server utilization high enough to offset rising infrastructure costs; absent that, the multiple risk is to the downside even if revenue remains strong.