JPMorgan says AI will need ‘astronomical’ amount of capex

JPMorgan revealed an almost astonishing amount of resources that will be necessary to deliver on the promises being made with artificial intelligence.

“The scale of demand for compute remains astronomical, with actual growth somewhat constrained by physical limitations,” the researchers note.

In addition to “astronomical,” the researchers also use adjectives like "extraordinary," “daunting,” and “staggering” in its 58-page report to describe what it’s going to take to make all of this work.

The report said that it will cost over $5 trillion, and up to $7 trillion, to build out the global data center and AI infrastructure.

And the funding will have to come from every public capital market, as well as private credit, alternative capital providers and governments.

According to the report, it will take 122 GW of global data center infrastructure capacity installations from 2026-2030, “at a rapidly accelerating rate.”

The problem is that the research team points out that of all the constraints facing the industry to pull this off, “the most important of those constraints is power.”

That’s because the current lead times for new natural gas turbines “have ballooned” over the last several years. And it has also historically taken more than a decade to build nuclear plants.

“Recent comments from certain power producers suggest some flexibility is possible from ramping up peakers more aggressively, but that has implications for retail prices,” the report states.

“Balancing ultimate retail electricity prices (still stable as a percentage of income, at least in the US) is a politically important and sensitive aspect of managing the data center boom.”

The researchers estimate that the annual data center funding needs in 2026 will reach about $700 billion annually, which can likely be covered by hyperscalers’ cash flow.

However, that number will reach $1.4 trillion by 2030, which is when alternative funding sources are going to have to be sought.

But the cash flow from hyperscalers “will have to do an enormous amount of the heavy lifting” regardless of other funding sources.

Revenue will have to be enormous for even modest returns

The researchers make the argument that private credit and other alternative structures will likely play a substantial role because of the flexibility “to better match cash flows and solve for ratings outcomes.”

And while they expect the banking system will “continue to be an important source of temporary/bridge capital,” they add that “financing long-term assets permanently with short-duration bank loans would be a material asset/liability mismatch.”

There has already been research written calling into question whether hyperscalers will ever be able to generate enough revenue to ever make the enormous amount of money being poured into building data centers pay off.

But JPMorgan’s research team puts this into even starker terms.

“Big picture, to drive a 10% return on our modeled AI investments through 2030 would require ~$650 billion of annual revenue into perpetuity, which is an astonishingly large number,” the researchers wrote.

This caught the eye of Ross Hendricks, an analyst for independent research firm Porter & Co., who has been skeptical about the amount of money being poured into AI.

“The AI math ain’t mathin,” Hendricks said in a post on X about that revenue number. “Two ways to fix it: 20x revenue in the next five years, or slash data center capex. I’ll take the latter on 100 to 1 odds.”

But if there were a pullback in capex, Hendricks added, it would mean “hyperscaler cloud computing earnings and NVDA chip sales evaporate…and with it, the entire stock market.”

The researchers note that it will ultimately take a “long-term debate” about how the funding is spread across corporations, governments and consumers.

“Regardless, even if everything works, there will be (continued) spectacular winners, and probably some equally spectacular losers as well given the amount of capital involved and winner takes all nature of portions of the AI ecosystem,” they wrote.