AI Spending Is Booming. The Economy Is Not Growing to Match, and the Gap Has a Mechanism

The puzzle

The spending figures are hard to square with the growth figures, and that is the puzzle J.P. Morgan sets out. Real US GDP grew at a 2.2 percent annualised rate in the second quarter of 2026, and has averaged 2.3 percent since the start of 2025, even as companies pour money into AI hardware and infrastructure. m:…1215

ABN AMRO frames the scale differently, grouping AI with defence and clean energy into a capex troika set to reach 3.5 trillion dollars in 2026 across the US, Europe and China, or about 4.5 percent of their combined GDP. EU capex alone is projected at 725 billion dollars in 2026 and 793 billion in 2027. Spending on this scale should, on a simple reading, move the growth needle further than it has. m:…0373

Markets, for their part, are still rewarding the story. Emirates NBD notes the Nasdaq rose 1.3 percent on Tuesday and the S and P 500 gained 0.6 percent, while South Korea's KOSPI rose around 6 percent on Wednesday and consumer confidence rebounded to 91.7 in October from a revised 84.2 in September. The rally and the growth data are telling different stories, and both can be true at once. m:…6446

Where the spending leaks

J.P. Morgan offers a mechanical explanation. Much of the hardware spending for AI is imported, which subtracts from US GDP, and net imports of computer hardware have risen sharply to nearly match the level of investment. Annualised private investment in computers and peripheral equipment has risen to roughly 365 billion dollars since mid-2023, so a large share of the headline boom never lands in domestic output. m:…1215

A second leak is displacement. Business investment in information processing equipment and software has risen by about 1 percentage point of GDP since early 2023, J.P. Morgan notes, while investment in other parts of the economy has fallen by about 0.8 percentage points over the same period. AI spending is crowding out investment elsewhere, which means the net addition to growth is smaller than the gross AI figure suggests. m:…1215

ABN AMRO adds a supply-chain twist: Taiwan and Korea source 62 percent of net trade in AI-related computing hardware, and the US and China account for more than 75 percent of net imports of that hardware. The buildout is global in its supply chain even when it is counted as domestic spending, which helps explain why the GDP effect is muted in any single country. m:…0373

A slower, still large cycle

None of this points to an imminent bust, according to ABN AMRO, which says an immediate AI capex bust is not expected but warns that long-term commercial viability remains uncertain because current returns do not yet justify the buildout. AI capex growth is projected to slow from over 70 percent in recent years to around 40 percent in 2027, while global investment growth across the troika slows from 22 percent to 20 percent. m:…0373

Northern Trust tells a similar story with different numbers, projecting AI capex growth falling from 54 percent next year to 12 percent in 2028, while hyperscaler AI infrastructure spending still reaches 1.2 trillion dollars in 2027 and 1.4 trillion in 2028. A decelerating growth rate on a much larger base is still an enormous absolute number, which is why the slowdown and the boom can coexist. m:…8435

The risk, Northern Trust warns, sits with weaker borrowers rather than with the hyperscalers themselves. Neocloud providers lacking strong balance sheets face refinancing risks as credit spreads widen, and bonds trading at spreads above 1,000 basis points reached 73 billion dollars in September 2026. The 10-year US Treasury yield stood near 5.1 percent by mid-October 2026, which raises the bar for any borrower without a strong balance sheet. m:…8435

The longer view is more optimistic. J.P. Morgan Wealth Management expects a 50 percent increase in productivity over the next 10 to 15 years from AI adoption, and ABN AMRO expects rising US defence spending to offset the moderating contribution of AI investment to GDP growth. This suggests the right question is not whether the spending is real, but when its productivity payoff arrives and who survives the financing gap in between. m:…1215 m:…0373