Two years ago, the standard line on AI capital spending was that it couldn’t possibly be sustainable, too much money going out, too little revenue to show for it, a bubble waiting for its moment. That line is getting harder to make with a straight face. Microsoft, Amazon, Google, and Meta are on track to spend somewhere between $740 and $770 billion combined this year on AI infrastructure, nearly double what they spent the year before, and the market’s reaction has shifted from alarm to something closer to acceptance.
I think the skepticism was reasonable in 2024 and is increasingly wrong now, for a specific reason: the spending isn’t speculative in the way it used to be. Microsoft’s Azure is projected to hit $148.9 billion in the next fiscal year, up about 40% from roughly $106 billion this year. AWS is looking at $168 billion in net sales for 2026, up 30.7%, and it’s sitting on a $364 billion backlog of contracted future revenue that doesn’t even include the recent $100 billion Anthropic deal. That’s not capital chasing a bet about the future — that’s capital chasing demand that’s already been signed and billed.
Morgan Stanley’s own analysis backs this up, projecting 25 to 50% long-term returns on this generation of AI capex. The number that actually matters here isn’t the size of the spend, it’s the ratio between committed future revenue and the money currently going out the door. On that measure, this looks less like the fiber overbuild of 2000 and more like a handful of companies reinvesting real profit into demand they’ve already got under contract. This is the flip side of the AI chip selloff from earlier this month, which was the market getting nervous about the AI trade for a few trading sessions. This is the four companies actually building the infrastructure making a much longer, much more deliberate bet regardless of that short-term nervousness.
The genuine risk isn’t that this spending is irrational. It’s that it’s now load-bearing. Once $750 billion a year in AI infrastructure spending is baked into four companies’ growth stories simultaneously, any real slowdown in AI demand doesn’t just miss a quarterly projection, it undermines the entire thesis across the biggest companies in the market at the same time. That’s the actual bet being made here, not that AI demand grows, but that it grows in something close to a straight line, with no pause long enough to strand $770 billion in fresh infrastructure sitting half-used. Fortune’s reporting has the fuller breakdown of where investor patience is thinnest.
The math holds up today, and it’s held up for several quarters running now. Whether it holds up through an actual, ordinary economic slowdown, the kind that shows up every several years regardless of what’s driving growth at the time, is the question nobody in this story has had to answer yet. That’s not a knock on the spending. It’s just the part of the bet that hasn’t been tested.