AI adoption is slowing: ‘There’s no use case for frontier AI’ | Dean Baker

Summary

Dean Baker argues that the AI industry’s investment plans are far ahead of any demonstrated ability to generate enough revenue. He considers Gartner’s forecast plausible: the industry may fall short of the roughly $8.5 trillion in returns needed to justify infrastructure spending, and many cloud-based AI operations may disappear by 2030. Baker says the projected spending is enormous, while no clear customer base or use case appears capable of paying for it.

He acknowledges that AI has useful applications, but says modest subscriptions and business fees are unlikely to support the scale of investment being planned. In his view, there is particularly little evidence that expensive frontier models are substantially better than cheaper alternatives, including Chinese models, in ways that businesses would pay many times more for. He also points to limited evidence of broad productivity gains and says companies are beginning to question what they are getting for their AI spending.

Baker sees signs that growth may already be slowing. He cites a report showing AI spending was roughly flat in August and September, despite an industry that would need sustained double-digit monthly growth to meet its commitments. He argues that this raises serious questions about how OpenAI and Anthropic can fund the hundreds of billions of dollars in computing obligations they have reportedly taken on for coming years, especially if they remain unprofitable.

He warns that companies exposed to the build-out could face severe financial pressure. Baker points to high borrowing costs and falling share prices at firms such as CoreWeave and Oracle, and says independent “neocloud” providers are especially vulnerable. He expects some companies or assets may be acquired by larger cloud providers. Even if the industry merely breaks even, he argues, that would disappoint investors and could leave highly indebted companies unable to meet their obligations.

A major correction in AI investment could have painful short-term effects, Baker says: falling technology stock prices could reduce household wealth, weaken spending, and contribute to a recession. Over the longer term, however, he believes a slowdown could encourage businesses to adopt AI more carefully and focus on applications that produce measurable benefits. He expects meaningful advances in AI eventually, but doubts they will arrive soon enough to justify current investment plans.

Baker also questions arguments that stopping or delaying data-center construction would itself cause a recession. He says construction should proceed only where there is a genuine need for computing capacity, while environmental impacts and local opposition should be taken seriously. If projects are profitable, he argues, companies will have incentives to overcome obstacles; delays may instead reflect doubts about demand.

Finally, Baker says antitrust policy may become more important if struggling AI infrastructure firms are absorbed by a handful of hyperscalers. He does not claim consolidation would necessarily make existing tech-sector concentration much worse, but suggests policymakers could consider how to preserve competition among companies developing AI products.

Presenters and Contributors

• Isaac Pound — Interviewer and Tech Report host
• Dean Baker — Senior Fellow Emeritus, Center for Economic and Policy Research