By Matt Day and Ian King | Bloomberg
The largest builders of artificial intelligence data centers have doubled their debt load in the last five years, turning to borrowing to finance an unprecedented spending spree they claim is needed to transform the economy.
Alphabet, Amazon.com, Meta Platforms, Microsoft and Oracle, the five biggest spenders on new data centers in the US, collectively added some $350 billion to their debt obligations in the last five years, according to data compiled by Bloomberg.
Also see: California lawmakers seek to boost reserves ahead of hot AI IPOs
They’re betting heavily that cutting-edge AI services will mean a flood of new revenue down the line. Investors have enthusiastically backed the companies, snapping up new bonds issued in a variety of currencies. But buyers gave an unusually chilly reception this week to a $25 billion issuance from Amazon, people familiar with the matter told Bloomberg, a sign that there’s a limit to the amount of money available to back investment from the tech giants.
The cost of that borrowing is still relatively minor for most of the companies, which have been enormously profitable. Interest expense at the five topped $10 billion last year. That’s more than double where it stood in 2019, but it pales in comparison to the free cash flow of just one of them. Google’s cash from operations minus capital expenditure was $64 billion at the end of the March quarter.
Other companies balance sheets are showing more strain. Amazon’s free cash flow went negative in the quarter ending March 31. The cash burn at Oracle, whose debt stood at about 2.5 times its sales in 2025, is expected to accelerate. S&P Global Ratings on Thursday downgraded Oracle to the lowest investment-grade rating, citing the company’s growing AI spending.
Also see: Chip worker shortfall endangers US factory revival
Software companies tend to be high-margin businesses that don’t require much regular capital expenditure. For the industry’s largest players, that started to change with the advent of cloud computing, which required big investment in server farms. AI data centers, which are usually larger than prior facilities and feature pricier chips, supercharged that outlay.
“The nature of these businesses is changing very dramatically, and it’s changing abruptly,” said Gil Luria, an analyst with DA Davidson & Co. “That’s why their cash flow is so depressed right now.”
Luria said the companies argue that their projected return on investment in the form of new AI services — particularly stacked against relatively low interest rates they are paying on new debt — make the build-out worthwhile. “They’re telling us that,” he said. “But you can tell investors are not comfortable.”
Amazon Chief Executive Officer Andy Jassy said in April that he had “high confidence this will be monetized,” citing customer commitments to use much of the new data center capacity being brought online by his cloud unit, Amazon Web Services.
Also see: Heat adds to strains on areas with data centers, raising the temperature on AI debates
Meta CEO Mark Zuckerberg told Bloomberg in a recent interview that demand for AI computing power continues to outstrip supply. “So the main thing that that gives us a lot of confidence in is that continuing to build out this infrastructure is going to be a good investment,” he said.
Equity investors have taken an increasingly cautious view on how and when the largest cloud-computing companies — often called hyperscalers — will get a return on the massive outlay. Only Alphabet’s stock has outperformed the S&P 500 Index this year, and shares of Microsoft and Oracle have dropped more than 20%.
As the businesses gear up to report quarterly earnings beginning later this month, debt market investors will be focusing on spending plans that have become a bellwether for the health of the AI boom. The questions are expanding from whether the companies are keeping up with the expansion plans of their peers to how they’re funding it and when the payoff will arrive.
“I don’t know that we know whether Amazon, Google, Microsoft and Meta are actually going to get a return on investment on this,” said Jason Pompeii, a corporate debt analyst at Fitch Ratings. “It seems like a lot of demand hype that is very aspirational at this point.”
The largest hyperscalers have pledged to spend as much as $725 billion this year, primarily on data centers and the Nvidia Corp. chips that make them run. They’ve funded that spending with a combination of cash, new borrowings and, in Meta’s case, arrangements that keep some borrowing off the social media giant’s balance sheet.
A growing debt burden can leave companies less equipped to handle future crises or shifts in technology. Not even decades of dominance in technology make a company immune to the weight of debt, if that grip slips.
When Intel Corp. CEO Lip-Bu Tan took the top job at the chipmaker in 2025, he said the first order of business was to clean up its balance sheet to end what he later admitted were questions about whether the company could survive.
What was until 2022 the world’s biggest chipmaker, Intel had racked up debt under his predecessors, who were trying to fund investor paybacks, acquisitions and an ambitious production expansion. But as the borrowing continued, the company failed to field competitive AI chips, missing out on the massive opportunity that’s made Nvidia the world’s most valuable company. A bad choice on manufacturing technology resulted in products that ceded market share. Intel’s revenue began to shrink and soon the company reported what was once unthinkable: losses. Wall Street projects 2026 will be its third straight year in the red.
It took a bailout from the US government and investment from Nvidia, a company Intel had bullied for decades, to keep the Silicon Valley institution on a viable footing.
Hyperscale cloud companies are not anywhere near that point, DA Davidson’s Luria said. “This doesn’t look bad,” he said of their combined debt load. “If they were borrowing an order of magnitude more? That would look bad.”