Is an AI downturn unavoidable?

The AI boom continues to grow and is now among the largest of the tech-related booms over the past two centuries. Most other examples ended abruptly, often painfully, even where the technology delivered strong economic benefits. That risk looks significant for AI, too.

For the near term, the key issue is whether the strong performance of AI-related stocks and the massive data-centre buildout could give way to a sharp pullback. This Research Briefing focuses on this risk.

What you will learn:

• AI-related capex accounted for an estimated 27% of US fixed investment in Q2 and cumulative investment in the sector has topped 2% of US GDP in the past three years. Among historical tech booms, only the UK railway mania of the 1840s showed a larger surge in investment.
• The surge in investment reflects hopes for very large returns from this new technology, but the evidence for these is limited so far, with utilisation-adjusted measures of US productivity growth decelerating. Notably, the dotcom boom of a quarter of a century ago produced productivity benefits quite quickly, so AI will also need to result in strong tangible benefits soon, especially with AI firms now showing negative free cash flow.
• Current consensus estimates of profit growth in the sector look inconsistent with our estimates of the likely productivity gains from AI.
• Importantly, for there to be a sharp correction in the AI sector, it doesn’t require the technology to prove worthless, only for there to be disappointment relative to the currently elevated expectations. There are a number of possible triggers for a correction, including concerns over profitability, incoming negative news on productivity gains and demand, regulatory risks, and rising US interest rates. These risks are likely to intensify as we enter 2027.