Every time Morgan Stanley’s cloud team updates its spreadsheets, the number gets bigger. The bank now expects global cloud capital spending to reach $1.2 trillion in 2027, up 30% from a year earlier and $170 billion above its previous estimate, according to the latest forecast. The four biggest US cloud operators remain capacity-constrained, with demand running ahead of supply. The spending curve is still climbing, and the bank sees no near-term break.
The revised number reflects what the operators themselves are saying. Amazon, Microsoft, Google and Meta have all raised their infrastructure spending plans, and the pipeline of data center construction is longer than it has ever been. The bank’s analysts said the increases are driven by demand that continues to exceed what the companies can supply.
The gap between demand and supply is the engine of the forecast. External cloud customers are buying capacity, and internal AI workloads are consuming the rest, leaving the operators in the position of rationing what they have. That scarcity is what justifies the spending and, in the bank’s view, the valuations.
The $170 billion revision is a statement about momentum. Estimates of this size are usually revised in increments of tens of billions, and a jump of this scale in one update says the underlying buildout is moving faster than the models assumed. The bank’s own aggregate estimate runs even higher, above the consensus figure it tracks.
The revision path is worth reconstructing. Morgan Stanley raised its estimates after the operators reported their latest quarters, and each report confirmed that the buildout had not slowed. The pattern, higher with every earnings season, is part of what makes the forecast credible to clients.
The capacity constraints are the detail that matters most. When the largest operators cannot serve all the demand they see, the constraint becomes the pricing power, the queue and the allocation of scarce compute. Analysts said the constraint, not the demand, is what will determine how the buildout unfolds.
The forecast has a direct bearing on the AI valuation debate. Critics argue that AI spending is a bubble, with capital being poured into infrastructure that may not earn a return. Morgan Stanley’s answer is embedded in the numbers: the operators are still selling everything they can build.
The bank’s analysts said the risk that dominates the cycle, overbuilding, remains distant. Capacity is being absorbed as fast as it is created, and the backlogs are growing rather than shrinking. The signs of a peak, in their view, are not visible in the supply-demand balance.
The counter-arguments are familiar. Skeptics point to efficiency gains in models and software that could reduce the compute needed per task, and to the risk that some AI services never find enough paying customers. Morgan Stanley’s answer is that the operators’ own capacity constraints make those risks a problem for later, not now.
The energy question is the newest variable. Data centers at this scale consume electricity in quantities that strain local grids, and power availability is becoming the binding constraint on the buildout. The forecast’s trajectory assumes the energy can be found, and that assumption is now part of the market’s debate.
The concentration of the spending is worth noting. Four companies account for the bulk of the $1.2 trillion, which means the forecast is really about the decisions of a handful of CFOs. If any one of them flinches, the curve bends.
The forecast is conditional, and the conditions include capital costs and returns on AI products. If the returns disappoint, the CFOs will trim, and the curve will bend faster than any model predicts. The bank’s view is that the constraint today is supply, and the forecast follows supply until the evidence changes.
The equipment suppliers are the indirect beneficiaries. Chip makers, server builders, power companies and construction firms all sit downstream of the hyperscaler budgets, and the raised forecast ripples through their order books. The supply chain for AI is being sized to the new number.
For the operators themselves, the spending is both a bet and a requirement. They cannot afford to underbuild in a market where capacity is the product, and they cannot afford to overbuild in a market where returns are scrutinized. The forecast describes the balance they are trying to strike.
The comparison to earlier technology cycles cuts both ways. Telecom companies overbuilt fiber in the late 1990s and paid for it for a decade. But cloud capacity differs in one respect: the customers are paying for it now, not hoping to find them later.
The memory and component makers feel the forecast directly. Every data center built is a buyer of memory, power, networking gear and servers, and the raised numbers extend their order visibility by years. The forecast is why the supply chain keeps expanding capacity even as some investors question the demand.
Morgan Stanley’s track record on this forecast has been a series of upward revisions, and the new number extends the pattern. Until the supply-demand balance turns, the bank sees no reason to change the trajectory. The $1.2 trillion figure is the current truth of the AI buildout, and it may not be the last.
