Morgan Stanley analysts project that spending by the largest cloud companies on artificial-intelligence infrastructure will reach $1.2 trillion by 2027, according to a report circulated this week and cited by Readhub. The forecast covers capital spending by hyperscale operators including AWS, Google and Meta Platforms, and it describes a build-out that continues to accelerate even as investors debate how quickly the investments will pay off.
The figure is large enough to strain belief, and the bank’s analysts acknowledge the skepticism. The report argues that the spending is driven by demand that is already visible, including the scramble for data-center capacity, the shortage of advanced chips and the pricing power that suppliers have enjoyed. The numbers are an extrapolation of trends already in the data, the analysts say, not a prediction that companies will suddenly change their behavior.
Meta’s announcement on the same day that it would double its Louisiana data-center investment to $50 billion served as a footnote to the Morgan Stanley forecast. When a single company is committing $50 billion to a single campus, a $1.2 trillion industry total stops sounding theoretical. The report and the announcement together describe an industry that has moved past planning and into construction, with contracts signed, crews on site and power lines being run.
The forecast builds on numbers that are already large. The hyperscalers have been raising capital-expenditure guidance for several consecutive quarters, and the pace of announcements shows no sign of slowing: data-center campuses are being planned in states and countries that had no such projects a few years ago, and the supply chain, from power transformers to construction labor, is straining under the load. The bank’s analysts note that the trend has a self-reinforcing quality, because capacity attracts customers, and customers justify more capacity. That dynamic is what makes a $1.2 trillion figure plausible to the forecast’s authors, even as it strains the credulity of investors who remember earlier cycles of overbuilding.
The industry has heard the skepticism before and has answered with the same argument: the demand is here, the contracts are signed and the economics, while stretched, still work. The cloud companies point to the revenue their AI services generate, to the backlog of customers waiting for capacity and to the competitive necessity of building before rivals do. Each of those arguments has been made in earlier infrastructure booms, and each proved true for a while before the inevitable reckoning. The question the market cannot answer yet is whether this cycle’s reckoning will come, and what it will cost.
The question that follows the spending is the one investors keep asking: when do the returns arrive? The Morgan Stanley report spends much of its length on that question, and its answer is that the returns are already starting to show up in revenue for the cloud companies, though not yet in proportion to the capital deployed. The gap between spending and earnings is the central tension of the AI trade, and every forecast of continued spending keeps the gap open.
There are reasons to believe the spending is rational. AI services are generating real revenue, and the companies building the infrastructure have the cash flows and balance sheets to fund it. The hyperscalers are competing for customers and for talent, and capacity is a weapon in both fights. The report argues that being late is more costly than being early, and that the industry has internalized that lesson from earlier technology cycles.
There are also reasons to worry. The history of infrastructure build-outs is that they overshoot, and the history of technology spending is that it consolidates around a smaller number of winners than the initial wave suggests. Some of the capacity being built now will be underused if the demand does not materialize as expected, and the write-offs could be enormous. The Morgan Stanley report acknowledges the risk and argues that the asymmetry favors building anyway.
The forecast also has implications beyond the cloud companies. If hyperscalers spend $1.2 trillion by 2027, the suppliers will be the immediate beneficiaries: chip makers, power-equipment companies, data-center builders and the utilities that serve them. The investment thesis for a wide swath of the technology and industrial markets now rests on the same forecast, which is part of why the market reacts so sharply to any sign that the hyperscalers are hesitating.
For investors, the report is a roadmap of what to watch. The key indicators are not the forecasts themselves but the quarterly capital-spending guidance the cloud companies issue, the pace of data-center construction and the revenue growth of the AI services that are supposed to justify the spending. Each quarter either confirms the trajectory or begins to bend it, and the market prices the news accordingly.
The $1.2 trillion figure will be cited for the rest of the year as both a promise and a warning. To the companies building the infrastructure, it is the size of the opportunity. To the skeptics, it is the size of the risk. The forecast does not settle the argument; it sharpens it. The spending will happen, the report says, and the question of whether it earns a return will be answered in the earnings reports of the years to come.
