The AI boom may transform the economy and still destroy capital, with stranded data centres and heavy investor losses

Is AI crowding out other investments? Paul Krugman recently wrote that it is 'crowding out other potential uses of capital'. At first glance, data appear to support his argument. US hyperscalers like Amazon, Google and Microsoft are pouring unprecedented sums into AI with capex expected to reach $800 bn in 2026. Yet, overall investment has barely budged.

A Federal Reserve Bank of New York study estimates that gross domestic investment in the US accounted for 21.3% of GDP in the 5 yrs before Covid. Since 2022, when companies began investing heavily in AI, the share has edged up to just 21.5%.

Also read: Is the AI boom real or a bubble? Nikhil Kamath asks Silicon Valley founders


So, to conclude from these data that AI is crowding out other investments is misleading.

The argument assumes the economy has a limited pool of capital and every dollar invested in AI is a dollar unavailable to other sectors. For an open economy like the US, this assumption is highly questionable. Capital can flow in from abroad, and higher expected returns can attract additional financing rather than simply redirect existing funds.

The argument also rests on a distinction between AI and non-AI investments that becomes increasingly difficult to sustain if AI is, indeed, the transformative technology its proponents claim it to be. AI is not just another sector competing with healthcare, education, manufacturing and agriculture for investment dollars. Its promise lies precisely in its ability to transform these sectors.

If AI raises productivity across the economy, spending on AI cannot be neatly separated from investment in the industries that benefit from it. The distinction may make sense in accounting statements. But it becomes less useful when assessing the broader economic consequences of that spending.

Consider healthcare. If AI makes surgery more precise, reduces medical errors and enables better diagnoses, investment in AI would also be an investment in healthcare's productive capacity. Elon Musk has gone further, predicting that AI and humanoid robots will revolutionise medicine, making the best healthcare virtually free and superior to what even world leaders receive today.

Musk may be exaggerating. Whether or not his vision materialises, the underlying point stands: value of AI investment may ultimately be realised in sectors far removed from the companies developing the technology.

The same applies to education. If AI enables personalised instruction and improves learning outcomes, investment in AI is also an investment in education. If it helps farmers apply fertiliser more precisely, reducing waste and raising yields, it is contributing to agricultural productivity. And if it helps manufacturers produce more with fewer resources, AI spending is also an investment in manufacturing.

The real question, therefore, is not whether AI is crowding out non-AI investment, but whether AI can make investment across the economy more productive. If it does, today's apparent trade-off between AI and other sectors would be dead wrong. What looks like a diversion of investment from healthcare, education or agriculture could, instead, be the financing of technologies that transform those very sectors.

True, an open economy does not eliminate crowding out altogether. Even when capital is available globally, AI investment will compete for workers, land, electricity, construction capacity and other scarce resources. It can also raise financing costs if it increases demand for funds faster than the supply of savings responds.

Yet, nor does a constant investment-to-GDP ratio prove that AI has displaced other investment. That would require evidence about what investment would have been in its absence. But judging AI investment solely by how much it adds to aggregate investment risks missing the bigger picture. The ultimate test is not how much money goes into AI but how much value it creates beyond AI itself.

What is clearer, however, is that even if AI is not crowding out investment in other sectors, it could leave behind a great deal of obsolete investment. If past investment manias are any guide, the AI boom may end with only 3-4 leading companies emerging as durable winners.

Data centres built by companies that fail, along with the power-generation capacity and water infrastructure constructed to support them, could prove redundant. Some of these assets may find alternative uses. But others may become stranded or represent outright waste.

This is not merely a theoretical concern. Bank for International Settlements (BIS), in its 2026 report, warned that exuberance surrounding AI could lead to a prolonged investment bust. If AI fails to deliver on its promises, if its benefits accrue to only a narrow set of industries, or if its expansion imposes substantial environmental costs, consequences could be serious.

Also read: Big Tech earnings a test of the AI boom amid market turmoil

An investment boom built on exaggerated expectations could leave investors with heavy losses, and the economy with underused infrastructure, wasted resources and an environmental burden.

Yet, these risks do not negate AI's transformative potential. So, the real question is not simply whether AI is crowding out non-AI investment. It is whether the productivity gains AI ultimately delivers will justify the resources committed to it today. AI may transform the economy even as the AI investment boom destroys capital and imposes a huge environmental burden. The technology could be revolutionary, while the investment cycle surrounding it turns out to be a costly mistake.

(The writer is professor of social policy, Columbia University, US)