For years, the largest cloud companies paid for their data centers the way they paid for everything else: out of the cash their businesses threw off. That habit is breaking.
Alphabet, Amazon, Meta, Microsoft and Oracle, the five hyperscalers that dominate cloud computing, are on track to spend a combined $4.2 trillion on capital expenditures in the four years through 2029, according to FactSet. The figure is large enough that it has begun to change how the companies pay for their own growth.
The five once funded their buildouts almost entirely from operating cash. That has shifted over the past eighteen months. The share of capital spending financed by new debt climbed from 9 percent in fiscal 2024 to 32 percent in the latest twelve months, according to FactSet, and equity has returned to the funding mix after a long absence.
The clearest signal came in June, when Alphabet priced an $84.75 billion equity raise, the largest such transaction ever by a listed company, including a $10 billion private placement with Berkshire Hathaway. Alphabet said the money would support general corporate purposes including AI capital spending.
The debt markets have absorbed the rest. Amazon recently placed a $25 billion bond, and Oracle has said it plans to raise roughly $20 billion of debt in its next fiscal year as part of a $40 billion debt-and-equity program. Oracle raised $43 billion of debt and $5 billion of equity in its most recent fiscal year alone.
The reason is that spending has outrun the cash the businesses generate. Aggregate capital spending for the five companies is expected to top $690 billion in their respective fiscal 2026 periods, up more than 80 percent from a year earlier. Analysts at FactSet expect free cash flow to fall near zero, or turn negative, at every one of the five except Alphabet and Microsoft.
The trajectory is steep even by the industry’s own recent standards. Aggregate capital spending for the five companies has climbed from about $95 billion in fiscal 2020 to roughly $490 billion over the latest twelve months, according to FactSet, with the money directed at AI compute, data centers and the power to run them.
The sums reflect the front-loaded economics of artificial intelligence. The companies are buying land, power, networking gear and memory chips years before the revenue from AI workloads fully arrives. Higher component prices, particularly for memory, have pushed the bill higher even as the growth rate of spending peaks.
The companies keep raising their own targets. Alphabet lifted its 2026 capital spending guidance to $180 billion to $190 billion, up from $175 billion to $185 billion, and has signaled a further increase for 2027. Meta raised its own 2026 guidance to $125 billion to $145 billion, from $115 billion to $135 billion.
Other forecasters tell a similar story. UBS has estimated the hyperscalers will spend about $4.1 trillion on AI infrastructure from 2026 through 2028, more than three times what the group deployed over the previous six years. The three largest cloud companies are on track to spend more than their entire cloud revenue on capital expenditures this year, according to UBS.
Leasing has become a quieter part of the bill. Combined lease commitments across the five hyperscalers now stand near $820 billion, according to FactSet, obligations that sit off the balance sheet but bind the companies for years. Oracle is the most exposed, with lease terms running fifteen to nineteen years.
The financing mix carries consequences. Debt on this scale raises interest costs and, eventually, borrowing rates, while equity raises dilute existing shareholders. Alphabet’s decision to raise equity rather than borrow was read by analysts as an effort to protect its balance sheet and credit rating before markets grow crowded.
Analysts said the shift matters beyond the companies themselves. The hyperscalers now compete with governments for capital, and the debt they issue will compete with every other borrower for investor money over the next several years. With the initial public offerings of Anthropic and OpenAI expected, the capital markets face substantial new AI-related supply.
Investors have not fully reconciled themselves to the numbers. Some have pressed executives on when the buildout pays off, since the returns on AI are expected over a longer horizon while the costs are all up front. The companies’ answer, delivered on successive earnings calls, is that the risk of underbuilding is greater than the risk of overbuilding.
What the money buys is increasingly a matter of geography and physics as much as software. Data centers require land, cooling and, above all, power, and the five companies are now among the largest buyers of energy in the country, competing with utilities and other industrial customers for the same grid capacity.
The spending has also redrawn the industry’s pecking order. Companies once measured by software margins are now judged, in part, by how much concrete and power they can secure, and the capital markets are pricing them accordingly.
For now, none of the five has signaled it intends to slow down. The question is no longer whether the spending will continue, but who pays for it and on what terms. The answer is increasingly written in bond prospectuses and equity offerings rather than in operating cash.
