The danger confronting the world economy lies in the cumulative impact of shocks over the past five years. An economy can survive one shock and emerge less capable of absorbing the next. Households draw down savings, businesses lose margins and governments deplete their fiscal buffers. Eventually, a disturbance that would once have been manageable can become the trigger for a crisis.

The 2020s have repeatedly tested households’ capacity to recover. The wars in Ukraine and Iran have inflicted food and energy shocks and have raised living costs directly; higher prices for other commodities have fed through to manufactured goods. Since January 2022, consumer prices have risen roughly 20% in the US and Germany and 25% in Britain by September 2026. Lower inflation, whenever achieved, does not reverse those increases. Families whose earnings have failed to keep pace continue paying for yesterday’s shocks.

Weak growth makes recovery harder. Much of Europe has struggled to generate the income gains needed to restore purchasing power. On the latest readings, the conventional misery index, combining inflation and unemployment, exceeds ten in France, Italy and Spain and has reached a historically high level in Germany. Even that measure misses the cumulative strain: today’s inflation rate says little about savings already depleted or consumption repeatedly postponed.

AI is adding to this with uncertainty over future incomes. Its effects can appear through fewer recruits and unfilled vacancies as well as redundancies. Stanford research links AI exposure to weaker employment among young Americans, principally through reduced hiring; it does not establish economy-wide displacement. The distinction matters. A graduate unable to enter the labour market experiences the shock even without appearing in a layoff announcement. Productivity gains may arrive unevenly and with delays.

And, ironically, AI is creating a financial vulnerability of its own. Infrastructure expansion is transforming the funding requirements of technology companies once associated with abundant cash and modest borrowing. Alphabet’s latest reported long-term debt exceeds $125 billion; Meta has $84 billion of notes outstanding. These figures capture only part of the obligations supporting the investment boom.

Cloud purchases, data-centre leases and commitments to infrastructure providers extend the exposure beyond conventional borrowings. Meta alone reports more than $450 billion of non-cancellable contractual commitments. Such contracts are not interchangeable with debt, and adding obligations across counterparties can double-count the same investment. Nevertheless, they commit future cash flows. If demand disappoints, the resulting pressure can travel from AI developers to infrastructure owners and their financiers.

The distinction between established technology groups and newer AI specialists is crucial. Alphabet and Meta have substantial existing businesses supporting their spending. Newer developers must demonstrate that expanding revenues can eventually cover enormous computing bills and financing requirements. Revenue growth is not synonymous with cash generation, particularly when serving each additional customer carries substantial costs.

Private credit adds another channel of vulnerability. As I argued previously, opacity, leverage and connections with banks and institutional investors complicate assessments of where losses ultimately reside. An AI investment reversal could therefore coincide with deteriorating corporate loans and weaker household demand. Different portfolios may prove exposed to the same underlying deterioration, undermining diversification precisely when it is needed.

None of this makes a crisis inevitable. The IMF’s latest assessment finds banks broadly resilient, although weaker institutions remain vulnerable. The question is how far that resilience extends when shocks reinforce one another. Unemployment can weaken consumption and loan repayments; falling revenues can strain investment commitments; tighter credit can then deepen the slowdown.

Banks and governments should test that sequence explicitly. Stress-testing exercises need to combine employment losses, persistent inflation, disappointing AI returns and pressure on private credit, tracing common exposures and contingent funding calls. Governments must assess whether they can support vulnerable households while facing higher debt-service costs and competing spending demands. Central banks may confront inflation that limits their freedom to ease.

The next shock’s severity will depend on its size, its connections and the buffers remaining when it arrives. Preparing for that possibility means measuring the capacity left after successive shocks. Are banks and governments ready when the next shock meets an exhausted economy?