The surge in AI infrastructure spending: between growth and financial dizziness

Irrational exuberance. The boss of JPMorgan did not use this formula. But the dizziness she describes clearly hovers above her latest figures. After the stock market panic that cost Michael Dell $15 billion in a single session, Jamie Dimon has just put a coin back into the worry machine, with a dizzying estimate.

From 300 to 700, soon 1,000 billion dollars

Guest of JPMorgan's 11th annual conference in India on Sunday, the CEO delivered an edifying trajectory. Investment spending by cloud giants (the famous hyperscalers, those companies whose infrastructure serves as the backbone of the entire sector) increased from around $300 billion last year to nearly $700 billion this year, according to comments reported by CNBC. They could cross the 1,000 billion mark in 2027.

Dimon sees this rush as a real engine of growth: around one point of additional GDP each year. The downside is inflation, which he judges likely to rise as companies hire, build factories and buy equipment at high prices. On the return on investment, he refuses to decide directly. “Sometimes it’s just the price to pay to stay in the race,” he summed up, preferring to speak of a mandatory investment rather than a calculated bet.

UBS thinks even bigger than Dimon

And Dimon would almost be optimistic. The bank UBS projects a total of 4,100 billion dollars in cumulative AI infrastructure spending between 2026 and 2028, including 1,447 billion for 2027 alone, more than the trillion mentioned by the boss of JPMorgan. Amazon, Alphabet and Microsoft alone would devote 102% of their cloud revenues to these investments this year, a figure which does not mean that they spend more than they earn, but that all of this windfall immediately goes back to data centers and chips.

None of this guarantees a return. The difference between the winners and losers in this race will not come down to who spends the most, but who manages to transform these billions into recurring revenue. A bet, not a certainty.