The global economy's figures this year contain a key puzzle. Theoretically, 2026 was expected to be the year that global trade wars finally left a dent in global growth figures. Goods trade grew by 4.6% in 2025 and the WTO's March 2026 outlook predicts that growth will slow to 1.9% this year,1 due to the lagged impact of tariffs on the data, alongside the conflict-driven uncertainty in the Middle East. Contrary to expectations, global growth forecasts have proven far more resilient than the trade backdrop itself would indicate.

The Outlier: AI Capex

A single category of investment explains most of the divergence, with the WTO's March 2026 outlook attributing close to half of 2025's goods trade growth to AI-related investment alone,2 expenditure that has had a countercyclical trajectory relative to broader trends in the global economy. The WTO estimates that AI-related spending accounted for around 70% of total investment growth in North America over the first three quarters of 2025.2 This is an incredibly high concentration in a single investment class with few parallels in recent trade history.

The scale involved extends beyond the technology sector with effects extending into semiconductor supply chains, construction and power generation. This is unsurprising once the underlying spending is considered; the four largest US hyperscalers, Amazon (AWS), Microsoft (Azure), Google (Google Cloud) and Meta, are forecasted towards a combined capital expenditure of $725bn in 2026, a 77 percent increase from roughly $410bn in 2025.3 While trade flows are contracting due to tariffs and other firms are visibly withholding investment until policy conditions clarify, this category of spending is expanding, seemingly decoupled from the broader trade and policy environment.

AI hyperscaler capex vs. global goods trade volume, indexed to 2024 = 100. Source: company earnings releases; World Trade Organization.

2024 index baseline: full-year capital expenditure disclosed by each company on its earnings call.4

The Mechanism

The three aforementioned channels which are semiconductors, construction and power, are where this expenditure is realised. At this scale, an order for AI infrastructure is fundamentally an order for physical goods: advanced chips from Taiwan and Korea, specialised construction contracts and turbines and transformers sourced from suppliers which have no direct exposure to the AI industry itself. The WTO credits this spending as supporting wider trade figures rather than being a domestic phenomenon: the capital is mostly American in origin, but the supply chain it activates is not.

An additional channel is via financial markets rather than physical supply chains. AI-exposed firms have experienced equity gains which have supported household and corporate wealth well beyond the technology sector, stimulating a confidence effect that extends into spending decisions. Another significant development involves the financing of this spending. Historically, hyperscalers used retained earnings to fund capital expenditure but have now shifted towards debt, with more than $100bn issued by mid-March 2026 alone, already surpassing the $80bn across 2025.5 The transition from a self-funded corporate decision to a claim on credit markets is a shift whose implications remain underexamined.

The Fragility of Narrow Growth

Despite the clear value that it offers and its clear productivity potential, the growth which AI is currently resting on is more narrowly based than it appears. A structural concern is that previously growth has been dispersed across many countries, sectors and investment forms, however, it is now heavily reliant on an oligopolistic channel and financed increasingly by debt. Regardless of the strength of the underlying technology, this is a more fragile basis for growth; it is highly dependent on a few capital allocation decisions moving in a certain direction.

This fragility can be represented by a physical constraint. Data centre electricity demand rose by 17 percent in 2025,6 and the International Energy Agency expects global consumption from data centres to roughly double by 2030.6 A bottleneck is created via more than 2,500 gigawatts of projects remaining stalled in grid connection queues worldwide,7 which cannot be cleared by capital alone on a conventional construction timeline. In contrast to a financing shortfall, an electricity shortage is not possible to refinance away. A solution can be found in the long run as grid capacity can expand, but not at the current pace spending assumes.

The second constraint surrounds monetisation. Sequoia Capital has estimated that the industry now requires roughly $3 trillion in annual revenue to justify 2026's ~$1.5 trillion in infrastructure spending, a gap that has continued to widen through the year.8 Against that gap, however, cloud AI revenues at the major hyperscalers continue to grow rapidly while enterprise adoption is only just beginning, providing a reason to believe the gap may close faster than the raw numbers suggest. The key uncertainty behind this growth is whether this happens in time.

What This Growth Depends On

The uncertainty is already beginning to resolve itself, with some promising signs. Enterprise AI revenue at the major cloud providers has been rapidly increasing rather than decelerating through 2026 which suggests that the gap between monetisation and scale of infrastructure is closing.8 Grid connection queues remain a constraint, however hyperscalers and utilities are responding via direct power-purchase agreements, on site generation and increasing utility capital spending, which suggests the physical bottleneck is being addressed. Credit markets have continued to absorb AI-linked debt issuance without a sign of the repricing which would indicate lenders have lost confidence in the sector.5

These signals demonstrate that this is no longer just a matter of corporate spending decisions in isolation, but of where a meaningful proportion of global aggregate demand is now generated. With rapid advancements in AI, the most poignant question is whether this rapid technology growth will be met with growth of the global economy's capacity in tandem. If the same concentration that facilitated this growth is not managed carefully, it could become the source of its fragility. If its capacity does grow alongside it, this would allow the AI boom to be remembered as an investment that carried global growth through a difficult growth period.

Footnotes

  1. World Trade Organization, Middle East Conflict Weighs Further on Slowing Trade Outlook (opens in a new tab), 19th March 2026. ↩

  2. World Trade Organization, Global Trade Outlook and Statistics, March 2026 (opens in a new tab), 19th March 2026. ↩ ↩2

  3. Financial Times compilation of the four hyperscalers' Q1 2026 earnings, reported in Tom's Hardware, Google, Microsoft, Meta, and Amazon capex spending to hit $725 billion in 2026, up 77% from last year (opens in a new tab), 30th April 2026. ↩

  4. Statista, Big Tech's AI Spending to Reach $725 Billion in 2026 (opens in a new tab), 30th April 2026, tracking full-year capital expenditure disclosed by Amazon, Microsoft, Alphabet and Meta. ↩

  5. Allianz Research, AI Capex Cycle: War-Proof for Now (opens in a new tab), 25th March 2026. ↩ ↩2

  6. International Energy Agency, Data Centre Electricity Use Surged in 2025, Even With Tightening Bottlenecks Driving a Scramble for Solutions (opens in a new tab), 16th April 2026. ↩ ↩2

  7. International Energy Agency, Electricity 2026: Executive Summary (opens in a new tab), 30th January 2026. ↩

  8. David Cahn, Sequoia Capital, as reported in TechCrunch, Can AI Answer the $3 Trillion Question? (opens in a new tab), 9th July 2026. ↩ ↩2