AI data center boom tests the limits of hyperscaler spending

Hyperscalers including Alphabet, Microsoft, Amazon, Meta and Oracle are committing historic sums to AI data centers, but economists warn the spending depends on rapid revenue growth and broad productivity gains that are not yet visible. Jessica Wachter of Wharton estimates AI companies would need to increase their own productivity by a factor of 2.7 to break even by 2030, assuming the cost of capital, a 15% return and depreciation.

The pressure is rising as companies borrow more, pushing risk beyond shareholders into lenders, private credit funds, insurers and pensions. Data centers also face fast depreciation because GPUs, which account for some 60% of costs, are improving roughly every two years or so, potentially turning older facilities into stranded assets unless owners keep investing.

Meta’s Hyperion project in Louisiana illustrates the local exposure. The financing involves Blue Owl Capital, a joint venture, lease structures and guarantees, while Entergy says it has a 20-year guarantee from Meta for power purchases. Consumer advocates worry residents could face costs if the facility needs less power than expected or leases end early.

Gary Gensler and other economists expect a retrenchment at some point, even if AI itself continues to advance after a financial reset. The central question is whether frontier models and massive facilities can produce sustainable value before cheaper models, public opposition, debt costs and infrastructure commitments weaken the investment case.