Big Tech's AI bet: 5-year GPUs, 30-year buildings, and a 2027 refinancing clock

Big Tech is projecting to spend more than $700 billion on AI infrastructure in 2026 alone, with analysts estimating that off-balance-sheet AI-linked liabilities across the group have grown to roughly $1.7 trillion up from roughly $200 billion four years ago (TheNextWeb's read on the filings, Biggo's aggregation). The math analysts are using says the buildout has to earn about $2 trillion a year in new revenue by the end of the decade to clear a 10% return, the S&P 500's long-run average, on top of what the underlying businesses already produce. That is the size of the U.S. individual income tax take, and it has to come from a product category that did not exist in 2022.

Alphabet gave investors the first hard datapoint of the year on July 22. The company posted its first-ever quarterly cash burn of $5.9 billion on revenue of $119.8 billion, up 24% year over year. Operating income rose 30% to $40.8 billion. Google Cloud jumped 82% to $24.8 billion. The Q2 capital expenditure line more than doubled to $44.9 billion, and Alphabet raised full-year 2026 CapEx guidance to $195-205 billion, roughly $15 billion above the prior range. Sundar Pichai told analysts on the call that 2027 spending will "increase significantly" and framed AI demand as the "very early innings of what feels like a secular shift" (Biggo's call summary; the official numbers are in the Q2 10-Q exhibit).

The capex bill is now larger than operating profit. Alphabet's $40.8 billion in operating profit could not cover a $44.9 billion capex line, even before the company's working-capital and dividend obligations. The shortfall is being plugged with debt and cash on hand. The same pattern is showing up across the rest of the group: combined Big Tech AI capex is projected to clear $700 billion in 2026, and operating cash flow is no longer keeping up, which is why the financing has shifted from internally funded capex to debt and equity issuance.

Underneath the capex line is an asset-lifecycle mismatch the financial structures obscure. A modern AI training chip has an economic life of about five years before a newer generation makes it uneconomic to run. The data center, the substation, the cooling plant, and the long-term power purchase agreement have a 30- to 50-year life. The two are being financed as if they were one asset. Bondholders are being asked to lend against a 30-year building whose main engine depreciates in five. The math works only if the GPU is replaced before it loses value, and the building keeps earning on the next generation.

For the math to clear, two things have to hold. The first is the revenue line: AI products have to add roughly $2 trillion in annual revenue by 2030 across the handful of cloud giants funding the buildout, on top of the search, advertising, cloud, and consumer subscriptions that already exist. The second is the refinancing window. A large share of the AI-linked debt issued in 2024 and 2025 matures in 2027 and 2028, and the rates on the rollover will reflect whatever the lenders think of the cash flow by then. The 2026 numbers do not have to clear; the 2027 rollover does.

The bet is not crazy. Inference costs have fallen sharply over the last two years, which means the revenue side of the equation can keep growing even as the cost of serving each query drops. Secondary markets for AI GPUs, which barely existed two years ago, now give the capex an actual residual value. The incumbents have an option the challengers do not: if the demand curve flattens, they can outlast the smaller labs by running on the older silicon at lower margins. Incumbents have used this playbook before, absorbing smaller competitors on older infrastructure when the demand curve flattens, and the AI buildout is the next iteration.

Microsoft, Meta, and Amazon report within seven days, and three watch items apply to all of them: how each company describes the capex-to-free-cash-flow gap, what it says about refinancing language in the 10-Q, and whether any of them cuts the capex range rather than raising it. Alphabet's $5.9 billion burn is the first datapoint, not the verdict.

Off-balance-sheet financing is where the printed numbers run out. The $1.7 trillion off-balance-sheet estimate is an analyst synthesis, not a company disclosure, and the direction of travel is the point: a growing share of the bet sits outside the consolidated balance sheet, on structures whose accounting treatment was last controversial during the Enron era. The 2027 rollover will price that exposure whether or not the earnings calls name it.