AI Infrastructure Growth Will Slow by 2028

Financial forecasts indicate a shift from hardware spending to application efficiency as power constraints tighten.

Updated on Sept. 28, 2026 in Data Centers

Bold flat-color editorial illustration showing a single industrial transformer in a sparse landscape, representing systemic energy constraints. — Morgan Stanley forecasts a significant cooling in AI infrastructure investment as power grid constraints and regulatory hurdles stall hundreds of billions in U.S. projects. AI Illustration. Upload story photo >

Live Poll

Do you believe the long-term benefits of artificial intelligence justify the massive energy and infrastructure requirements?

Morgan Stanley projects that hyperscaler capital expenditure growth will drop from 93 percent in 2026 to 14 percent by 2028. This forecast identifies a 57 gigawatt power shortfall as the primary barrier to continued expansion in the United States.

Why it matters

The investment cycle is pivoting from heavy hardware procurement toward software and application development as developers encounter physical infrastructure limits. Power shortages and regulatory hurdles have already sidelined USD 286 billion in data-center projects since 2025.

The projected capital expenditure growth for cloud providers will fall to 61 percent in 2027 and 14 percent by 2028, down from the 93 percent peak recorded in 2026. This slowdown is complicated by a 57 gigawatt power deficit, representing a significant portion of current demand.

The players

Morgan Stanley

A multinational financial services firm that tracks global technology infrastructure spending trends and capital allocation shifts.

The details

Infrastructure development is stalling due to a convergence of power grid limitations, labor shortages, and regulatory bottlenecks that have halted or delayed USD 286 billion in U.S. data-center projects. As companies shift focus from raw hardware acquisition—the purchase of server chips and facilities—to software-defined efficiency, they target a 100 basis point (one percentage point) increase in net margins. This adjustment follows reports that AI adopters achieved an average 9.6 percent net productivity gain over the past year.

Since 2025, USD 286 billion in data-center projects have been delayed or cancelled.
In 2026, hyperscaler and neocloud capital expenditure growth reached 93 percent.
A 57 gigawatt power shortfall is projected between 2026 and 2028.
Capital expenditure growth is projected to fall to 61 percent in 2027.
Capital expenditure growth is projected to fall to 14 percent in 2028.

The Tech Race

This projection aligns with broader industry data indicating that the AI infrastructure race is shifting from an uncontrolled hardware arms race to a constrained capacity management phase. The 57 gigawatt shortfall acts as a definitive benchmark for how grid capacity will dictate the velocity of future deployments.

The shift toward software efficiency suggests that AI-powered applications may see improved performance optimization rather than merely increasing total computational volume. Users should anticipate this transition to manifest as more stable and efficient enterprise software tools as companies prioritize margin expansion over pure infrastructure expansion.

The takeaway

The transition from hardware-intensive spending to software-based margin expansion marks a maturing phase for the AI industry. Investors and developers should monitor the projected 2028 capital expenditure data points as the primary indicator for when these infrastructure constraints will settle into a new equilibrium.