The biggest AI spenders are slowing down just as forecasts pass $1 trillion

The slowdown at the top

Ramp's lead economist reports that the top 1 percent of businesses drive 80 percent of total AI spending, and that this group is starting to slow its spending on AI, according to a podcast discussion of Ramp platform data. m:…8241

The concentration is stark: the median firm on the Ramp platform spends 15 dollars per employee per month on AI, while the top 1 percent spend between 7,000 and 10,000 dollars per employee per month. m:…8241

Two features of the current spending base make a slowdown matter more than the median suggests. The Ramp discussion notes that 70 to 80 percent of revenues at AI companies come from other AI companies funded by venture capital, and that if venture-backed companies cannot raise their next funding round, their AI spending will stop. m:…8241

Forecasts still point up

UBS has moved in the opposite direction from the spending data, raising its aggregate AI capital-expenditure forecasts to 1.0 trillion dollars for 2026 and 1.4 trillion dollars for 2027. m:…0110

Janus Henderson puts the hyperscaler part of that buildout even higher in the near term, expecting hyperscaler capital expenditure to balloon to more than 1.1 trillion dollars in 2027, and estimating that capital spending on the AI buildout could average around 3.6 percent of GDP annually between now and 2032. m:…0851

Part of the tension may resolve through prices rather than volumes: The Ramp discussion notes that AI model prices have fallen by almost half in recent months, and the Baillie Gifford discussion notes that prices of tokens charged to use AI models are already falling. Cheaper models mean slower spending growth does not necessarily mean slower usage growth. m:…8241 m:…9609

The bet is spreading into bonds

Janus Henderson warns that the AI buildout is increasingly financed through investment-grade credit, so passive bond investors risk accumulating greater exposure to the same AI theme that already dominates equity markets. m:…0851

The firm puts numbers on that concentration: the S&P 500 carries AI exposure of 40 percent, the technology sector represents around 10 percent of investment-grade indices today and is expected to reach more than 20 percent of the iBoxx index by 2028, while hyperscaler duration-times-spread already rivals that of the six biggest US banks. m:…0851

Janus Henderson also notes that AI ecosystem spreads trade wide by around 35 basis points compared with the investment-grade average, and argues that active credit selection can limit exposure to hyperscalers while targeting issuers engaged in deleveraging. m:…0851

An old warning, applied carefully

In a Baillie Gifford discussion of the 1873 crash, historian Liaquat Ahamed argues that AI infrastructure spending is currently in the same zone as railway spending in the 1870s, and that equity markets are currently valued at approximately 300 percent of global GDP. m:…9609

The historical parallel comes with a sharp detail: half of US railway companies defaulted within five years of the 1873 crash, according to the same discussion. That is a warning about what can follow a buildout financed on optimistic assumptions, not a prediction that today's builders will fail. m:…9609

UBS offers the practical version of the same caution, advising investors to reduce single-stock concentration even as it raises its AI spending forecasts. The message across all four sources is consistent: the AI trade is now a concentration question, not just a growth question. m:…0110