How Big Tech’s $1tn AI Spending Spree Is Masking Global Economic Fragility

AI spending accounts for roughly 25-33 per cent of recent US GDP growth, while total US corporate capital expenditure is moving beyond $1tn this year. The IMF expects global growth of 3 per cent in 2026, down from 3.5 per cent last year, before accelerating to 3.4 per cent in 2027. The improved outlook compared with its April forecast is being driven directly by US-led technology investment.

That makes the AI boom more than a technology story. It has become a macroeconomic mechanism.

The investment cycle is doing the work of a stimulus

For much of the post-financial-crisis era, global growth was heavily influenced by monetary policy and the availability of cheap credit. The present cycle looks different. Private companies are now deploying capital on a scale capable of materially affecting national growth rates and international trade.

The hyperscalers are effectively building an industrial economy around computation.

Amazon is planning $200bn-$220bn of spending, including data centres and proprietary AI chips. Alphabet is allocating $195bn–$205bn to computing clusters and model training. Microsoft is budgeting $175bn–$190bn, including GPU infrastructure and long-term power arrangements. Meta is preparing to spend $130bn-$145bn.

The significance of those figures extends well beyond Silicon Valley. Every new data centre requires servers, advanced semiconductors, high-bandwidth memory, electrical equipment, cooling systems, copper, optical fibre and enormous quantities of electricity. Capital therefore moves through a long chain of suppliers before it becomes a digital service.

The effect is visible in Asian trade data. South Korean exports rose 63 per cent year-on-year in July, driven by demand for high-bandwidth memory, while the IMF raised its forecast for the country's 2026 growth to 2.6 per cent, 0.7 percentage points above its April projection. Taiwan's exports increased by roughly one-third. China's July exports rose 25 per cent and Japan's 22 per cent. Thailand and Malaysia are also outperforming IMF baseline projections as they participate in hardware integration and assembly.

This is why the current resilience can be misleading.

The world is not necessarily experiencing broad-based acceleration. A significant part of its growth is being generated by one exceptionally concentrated investment cycle.

Five balance sheets, global consequences

The financing of that cycle makes the dependence more consequential.

The hyperscalers are increasingly using corporate bond markets alongside equity and private capital. Bond issuance by the five leading companies is expected to reach about $250bn this year, roughly twice the previous level, with another near-doubling projected for 2027.

That introduces a new financial dimension to what was initially an infrastructure race.

The headline spending figures also do not capture every future obligation. Long-term power purchase agreements, unstarted data-centre leases and chip purchase commitments can sit outside conventional capital expenditure measures. Analysts cited in the briefing estimate that such exposure could reach as much as $3tn.

The issue is not that these companies are necessarily unable to finance their plans. They have enormous revenues, valuable assets and access to deep capital markets. The issue is what happens if the economic return on the investment arrives more slowly than the infrastructure spending.

Goldman Sachs' analysis of S&P 500 earnings calls provides an uncomfortable indication of the gap. Although corporate executives discuss AI applications frequently, only 11 per cent of companies have quantified particular AI use cases, 2 per cent have quantified a direct impact on earnings, and 7 per cent have disclosed identifiable AI-related operating expenses.

The technology can still prove transformative. But the macroeconomy is being asked to absorb enormous investment before the revenue case has been demonstrated consistently across corporate America.

That is the asymmetry at the centre of the boom.

Europe gets the machinery - and the risk

Europe occupies an awkward position in this new economic architecture.

It has important upstream suppliers. Companies such as ASML, Siemens and Schneider Electric benefit from the construction of data centres and the electrical infrastructure required to run them. Europe's ICT sector accounts directly for about 5.5–6 per cent of EU GDP, rising to 10–12 per cent when adjacent digitalised sectors, telecommunications and fintech are included.

But Europe lacks comparable consumer-facing hyperscalers.

Its position is therefore structurally different from that of the US. European industry can profit from supplying the physical machinery of the AI economy while European households, insurers and pension funds remain exposed to the financial performance of American technology companies.

That helps explain why the European Central Bank has been warning about the possibility of an eventual correction in AI megacap valuations. A fall in US technology shares would not remain confined to American investors if European financial institutions have significant direct and indirect exposure to them.

Norway offers an even clearer illustration of the concentration problem. Nicolai Tangen, head of Norges Bank Investment Management, which manages a $2.3tn sovereign wealth fund, has warned about portfolio risk as 10 technology companies account for roughly one-fifth of the fund's total value.

The concern is not simply that technology shares may fall. It is that the same companies have become simultaneously important to corporate investment, global trade, financial markets and sovereign portfolios.

A correction could therefore travel through several channels at once.

Computation has become an energy problem

The physical limits of the AI economy are becoming as important as its financial ones.

A data centre is not weightless software. It consumes electricity, requires transformers and electrical equipment, occupies industrial land and depends on vast quantities of specialised hardware. The resulting competition for grid capacity is already affecting wholesale electricity prices.

That creates an unusual monetary-policy problem.

Central banks can respond to weak consumer demand by lowering interest rates. They cannot easily manufacture additional transformers, copper, advanced memory or electricity by changing the policy rate. If AI infrastructure keeps bidding for those resources, industrial inflation can remain elevated even while other parts of the economy are weak.

The demand for chips, high-bandwidth memory, copper, optical fibre and liquid-cooling systems is therefore transmitting the AI boom into the real economy. Households and traditional businesses can end up paying more for electricity while non-AI companies compete for equipment, specialised workers and financing.

The result is a potential inversion of the usual technology narrative. Digitalisation was supposed to make economic activity more efficient and less resource-intensive. Frontier AI is making computation increasingly dependent on physical infrastructure on an industrial scale.

That can put central banks in a difficult position: growth is being supported by capital expenditure, but some of the same investment is sustaining the inflationary pressures that prevent rapid monetary easing.

The danger is not that AI fails

The historical comparison is revealing. Previous infrastructure booms, including the late-1990s telecommunications fibre buildout and earlier railway manias, created assets that ultimately proved economically useful while still producing periods of severe over-investment and financial correction.

AI could follow the same pattern.

The crucial question is therefore not whether artificial intelligence will have economic value. It almost certainly can. The harder question is whether the value arrives quickly enough, and at sufficient scale, to justify the capital being committed now.

If hyperscalers slow their spending because returns disappoint, the consequences would extend far beyond their shareholders. Asian semiconductor exporters would face weaker demand. European equipment suppliers would lose part of the investment impulse supporting their order books. Corporate bond markets would lose a major source of issuance. Energy and infrastructure projects built around anticipated data-centre demand could face a different economic equation.

Most importantly, the contribution of AI investment to US growth would diminish precisely because so much of current growth is tied to that investment.

That is the single-point-of-failure problem.

The global economy has found an extraordinarily powerful shock absorber at a moment when trade barriers, geopolitical conflict and expensive money are weighing on conventional sources of growth. But the absorber is being operated by a very small number of companies, increasingly through debt-financed capital spending, while markets and governments around the world become dependent on the continuation of their plans.

The deeper transformation is therefore not simply that AI is becoming an enormous industry. It is that private technology balance sheets have acquired a macroeconomic importance once associated more readily with governments and central banks.