Hyperscalers to Spend $4.1T on AI Through 2028, Recycling Cloud Revenue Entirely

The AI infrastructure boom is fundamentally reshaping how hyperscalers allocate capital. Rather than incrementally expanding data centers to match immediate demand, companies like Amazon, Alphabet, and Microsoft are building years ahead of expected usage — a strategy reflected in the fact that they will spend approximately 102% of their collective cloud revenue on capital expenditures in 2026. This ratio eases over time (to 99% in 2027 and 94% in 2028), but UBS's projections show that total spending continues to accelerate in absolute dollars, reaching $1.619 trillion annually by 2028.

What distinguishes this from past technology cycles is both the scale and the permanence. The $4.1 trillion projected through 2028 dwarfs the $1.292 trillion spent over the previous six years, and the trajectory suggests the industry is locking in a structurally higher capital base rather than pursuing a temporary upgrade. The spending is also distributed across a wider ecosystem: while Alphabet, Meta, Microsoft, and Amazon dominate, SpaceX, Oracle, and neocloud providers are increasingly significant contributors, reducing the risk that the buildout is a single company's speculative bet.

However, the investment thesis ultimately depends not on the spending itself but on whether AI demand grows large enough to keep the infrastructure utilized and profitable over years. Companies face a timing risk — infrastructure deployed today may take years to reach full utilization, and slower-than-expected demand growth could pressure margins. The winners will be those that convert this unprecedented capex into recurring revenue, high asset utilization, and durable free cash flow, not necessarily those spending the most.