AINEWS
Four of the five largest U.S. cloud providers — Microsoft, Amazon, Google, and Meta — are each committing more than $100 billion to AI infrastructure this year, and the question is whether AI revenue is growing fast enough to justify the bill.
Four of the five largest U.S. cloud providers — Microsoft, Amazon, Google, and Meta — are on track to spend roughly $602 billion on infrastructure in 2026, a 36% jump from 2025, according to a consensus of analyst estimates led by MUFG. Each of the four will individually commit more than $100 billion this year. The spending is a bet on AI demand, and it is large enough to reshape the industry's capital structure.
Roughly three-quarters of that bill, about $450 billion, is earmarked for AI infrastructure: the GPUs, the data centers that house them, the long-dated power contracts, and the networking gear that ties clusters together. The remainder is more traditional cloud capacity, where growth is steadier but the marginal dollar is less productive. The split is consistent across the cohort, with Introl and CreditSights landing on the same allocation. The share figure, not the headline total, defines what the buildout actually is.
The more useful comparison is to revenue. Capital intensity, the share of sales a company reinvests in infrastructure, is now running between 45% and 57% across the four. That range is historically unusual for software-adjacent businesses, and it places the cloud cohort closer to a regulated utility than to a software company. The bet is that AI demand is durable enough to absorb the spending. The question is whether the revenue line is keeping pace.
It is, but the gap is widening. Enterprise cloud infrastructure revenue is now north of $143 billion a quarter, with growth accelerating, according to The Register. Annualize that figure and it covers the $602 billion capex bill about one-to-one, except the cohort is also paying for everything else: operating costs, dividends, stock buybacks, and the ordinary capital maintenance of a cloud business that existed before any of this started. Operating cash flow does not cover the gap. Debt does.
The four raised roughly $108 billion in debt during 2025 to bridge the shortfall, and analysts project approximately $1.5 trillion in total debt issuance across the 2025–2028 buildout, per Introl. The cohort is expected to fund about half of global data-center spending over that period. The cost of carrying that debt is the second-order story the headline number hides, and it is the one that starts to matter when AI revenue conversion comes under scrutiny.
That scrutiny is starting. In June, Forbes reported that the gap between AI-related capex and AI-related revenue is widening, and that markets are beginning to reprice the names accordingly. The piece does not call the buildout a bubble, and it does not declare it a payoff. It points out that the discount rate the market applies to forward AI earnings has started to move.
The next checkpoint is Q3 2026 earnings, when all four companies will report capex guidance alongside any change in their AI revenue disclosures. If cloud revenue growth holds above 30% year-over-year, the capex narrative stays intact. If it slips, the $602 billion becomes the line item that gets questioned, not celebrated.