An ambitious call to build sustainable AI

This is less a “green tech” story than a unit-economics reset for AI. The market is still pricing AI as a software-margin expansion narrative, but the binding constraint may become power procurement, cooling density, and interconnect speed; that shifts value toward firms that can secure low-cost, reliable electricity and away from those whose AI spend is mostly an operating drag before monetization catches up. In the near term, the biggest risk is not that AI demand disappears, but that supply-side friction delays deployment and forces another leg higher in capex intensity, which can pressure FCF and multiple quality across the group.

Among the listed names, the relative winners are the hyperscalers with the broadest cloud pricing leverage and balance-sheet capacity to sign long-dated power deals. The weaker setup is the company whose AI spend is most front-loaded versus near-term revenue conversion, because higher electricity and cooling costs hit margins before any efficiency gains show up. A second-order effect is that efficiency improvements can actually accelerate usage growth through lower cost per inference, so the long-run revenue pool may expand even if near-term margins compress.

The contrarian point: the market may be overpricing a fast transition to nuclear/SMR and other “green AI” solutions. Those are structurally relevant but too slow for the next 4-8 quarters, so the tradable signal is still about who can finance, site, and power data centers today. Falsifiers are simple: if the next earnings cycle shows capex rising faster than AI revenue while guidance acknowledges power bottlenecks, the premium multiples should compress; if managements quantify better monetization per kWh and stable incremental margins, the thesis shifts back to bullish.