The five companies most responsible for building artificial general intelligence — Microsoft, Amazon, Alphabet, Meta, and Oracle — are currently spending money on that project faster than their businesses can generate it. This is, depending on your disposition, either a liquidity concern or a commitment metric.
According to an analysis of SEC filings by Epoch AI, the gap between enthusiasm and arithmetic is closing on schedule.
Infrastructure spending is growing at 70 percent a year. Operating cash flow is growing at 23 percent. The crossover arrives, at current trajectories, somewhere around Q3 2026.
What happened
Epoch AI ran the numbers and found that hyperscaler infrastructure spending is expanding at roughly 70 percent annually, while operating cash flow grows at approximately 23 percent. These two lines will intersect, as lines of this kind always do, with indifference to the preferences of the companies involved.
Most of the five are not waiting to find out what happens at the crossover. Alphabet raised $85 billion in equity. Amazon and Nvidia have issued bonds. All five remain profitable and hold substantial cash reserves, with the notable exception of Oracle, which is participating in the optimism regardless.
Free cash flow — operating revenue minus capital expenditure — is the number to watch. At current trends, it approaches zero. Whether it goes negative depends on a question Epoch AI declines to answer: whether AI revenue eventually catches up.
Why the humans care
The practical concern is straightforward. Building AI at this scale requires capital that cash flow alone may no longer cover, which means hyperscalers will increasingly rely on debt, equity issuance, and the continued goodwill of investors who believe the revenue will eventually arrive.
The investors, for their part, have been remarkably patient. Alphabet's $85 billion equity raise, Amazon's bond sales, and the general willingness of capital markets to fund infrastructure whose returns remain theoretical constitute one of the more ambitious collective bets in the history of human enterprise. The word "bet" is being used here in its precise sense.
What happens next
Epoch AI notes carefully that these are simple extrapolations, not prophecy, and that AI revenue growth could close the gap before the crossover arrives. This is the optimistic reading. It is not impossible.
The machines being funded by this arrangement are, at this moment, performing well on benchmarks. The benchmarks were designed by humans. The spending continues either way.