Tarek Hamid and his team of credit strategists at JPMorgan have been revising one forecast upward for seven months. On Wednesday, they did it again, lifting their estimate of what the world’s biggest technology companies will spend building out artificial intelligence to about $5.5 trillion by 2030.
The new figure, laid out in a research note this week, is roughly $400 billion higher than the bank’s November projection. JPMorgan’s strategists focus on hyperscale data center operators, the companies erecting the vast computing campuses that train and run frontier AI models. The revision reflects a simple observation: the buildout is getting bigger, and the bill is arriving faster than Wall Street expected.
The striking part of the forecast is not the headline number but how it gets financed. Of the $5.5 trillion, roughly $4.1 trillion will come from debt, the strategists wrote. That implies the hyperscalers will use borrowed money to cover a substantially larger share of their capital spending than they have so far, a shift with consequences for the corporate bond market, for interest rates, and for the credit ratings of some of the most valuable companies on earth.
The shift is already visible in issuance data. Since JPMorgan made its November forecast, companies have sold more than $300 billion of bonds tied to AI and data centers, according to the strategists. They described data center debt as the single biggest driver of the near-record pace of corporate bond issuance in the early months of this year. Investment-grade supply has flooded the market as operators pre-fund multi-year construction programs, and bankers say the pipeline shows no sign of slowing.
The economics help explain why. A single hyperscale campus can cost tens of billions of dollars, and the race to secure power, land, and chips means projects are often funded years before they produce a dollar of revenue. With cash flows still building, companies are turning to the debt markets to bridge the gap. JPMorgan’s strategists said market expectations have shifted: investors now assume the hyperscalers will carry meaningfully more debt in their capital structures than in prior cycles.
Amazon’s recent commitments illustrate the pattern. The company pledged $50 billion toward OpenAI’s record funding round, with $35 billion of that contingent on performance conditions, according to a breakdown of the deal disclosed in April. Amazon and its peers are simultaneously funding other companies’ AI ambitions and building their own infrastructure, and both activities consume capital at a pace that internal cash generation cannot match. Microsoft, Google, and Meta have each raised capital expenditure guidance this year, and each has signaled that data center construction will absorb the increase.
The borrowing spree is not without risk. Credit analysts point out that AI infrastructure has a long payback period and that the revenue attached to it remains concentrated among a handful of large customers. A slowdown in AI adoption, or a sudden shift in demand for compute, would hit an industry carrying tens of billions of new debt. Some investors also worry about crowding: if the hyperscalers keep issuing at the current clip, they will compete for the same buyers who fund the rest of the investment-grade market, and the ratings agencies will have to decide how much debt a company with $50 billion of annual capital spending can safely carry.
For now, though, the demand side shows few cracks. Bond buyers have absorbed the supply eagerly, and spreads on AI-linked issuers have stayed tight. The strategists noted that issuance linked to data centers helped push early-2026 corporate bond volume near record highs, a sign that the market’s appetite for AI exposure remains strong even as equity investors grow choosier about technology stocks. The equity market’s discomfort, in fact, is part of the story: with stock prices wobbling, companies that planned to fund expansion with equity have found debt cheaper and faster to arrange.
The $5.5 trillion figure also raises the stakes for the rest of the economy. Data centers are among the fastest-growing consumers of electricity in the United States, and utilities, turbine makers, and power developers are planning capacity around the hyperscalers’ disclosed roadmaps. Those plans, in turn, assume the funding JPMorgan is now forecasting. The forecast and the physical buildout have become mutually reinforcing, and both rest on the same premise: that the AI boom lasts long enough for the infrastructure to pay for itself.
JPMorgan’s note stops short of calling the spending rational or reckless. The strategists framed the increase as an accounting of what companies are already doing, not a prediction that they must. But the numbers carry an implicit warning. When a forecast of $5.5 trillion in spending, financed mostly with borrowed money, is revised upward within seven months, the market that funds it is betting that the world’s largest companies can keep turning debt into productive capacity faster than the bill comes due.
Analysts at other firms have been moving in the same direction. Several banks have lifted their AI infrastructure estimates through the first half of this year as hyperscaler capital expenditure guidance rose, and the figure is now a central variable in credit and equity models on both sides of the Atlantic. Few expect the next revision to go the other way, and that itself is the point: the debt market’s willingness to fund the buildout is what makes the buildout possible, and so far, the market has said yes.
