SKN | Morgan Stanley Raises U.S. Data-Center Power Shortfall Forecast to 33 GW as AI Infrastructure Demand Accelerates

Key Takeaways:

• Morgan Stanley raised its projected U.S. data-center power shortfall through 2028 to 33 GW after incorporating higher chip shipment estimates and accelerated power solutions.
• The bank now estimates a 57 GW gross power gap before additional generation solutions, up from its previous 38 GW forecast.
• Morgan Stanley increased its estimate for cumulative U.S. data-center power demand through 2028 to 97 GW, with accelerated measures expected to address roughly 24 GW of the gap.
• The bank’s revised model highlights how rack-scale AI systems are materially increasing electricity requirements across the data-center infrastructure chain.

Morgan Stanley has materially raised its forecast for the U.S. data-center power shortfall, reflecting a sharper assessment of the electricity requirements associated with the next generation of AI infrastructure. In its September 21 report, the investment bank estimated that the residual power gap could reach 33 GW by 2028 after accounting for onsite generation and other accelerated solutions.

The revision gives institutional investors a clearer indication of how Morgan Stanley is evaluating the infrastructure constraints surrounding continued AI deployment. The issue is no longer simply the number of chips being installed, but the amount of electricity required to operate increasingly dense computing systems.

Morgan Stanley Recalculates the Scale of the Power Constraint

Morgan Stanley now estimates a 57 GW gross shortfall through 2028 before additional power solutions, compared with its previous estimate of 38 GW. The bank arrives at that figure by comparing projected demand with 21 GW of data centers already under construction and 19 GW of available or contractable grid capacity.

The revised model places cumulative midpoint shortfalls at approximately 5 GW in 2026, 12 GW in 2027 and 33 GW in 2028. Morgan Stanley attributes much of the increase to higher chip shipment expectations and the industry’s transition from conventional 8-GPU servers toward integrated rack-scale architectures.

Higher Rack Power Requirements Change the Infrastructure Equation

A central part of Morgan Stanley’s revision is its reassessment of power consumption at the rack level. The bank now models an Nvidia Vera Rubin rack at approximately 234 kW, compared with 149 kW previously. For Rubin Ultra, the estimate rises to 600 kW from 415 kW.

These figures encompass more than GPU consumption. Morgan Stanley’s calculations include memory, networking, power delivery and liquid cooling, reflecting the increasingly integrated architecture of modern AI computing infrastructure.

The bank estimates that its revised demand forecast through 2028 is approximately 43% higher than its previous 68 GW projection. It also corrected earlier double counting involving certain powered-shell facilities, which had previously reduced the estimated residual shortfall.

Morgan Stanley Maps the Power Solutions

Morgan Stanley estimates that probability-weighted accelerated power measures could provide approximately 24 GW through 2028. The estimate includes additional behind-the-meter turbines and engines, fuel-cell capacity from Bloom Energy and data-center locations at operating nuclear facilities.

The bank emphasizes that these figures remain model estimates rather than utility-approved load or energized capacity. Its framework converts projected accelerator shipments into active fleets, rack requirements and facility-level electricity demand, incorporating assumptions around utilization, deployment schedules, chip retirements and power usage effectiveness.

Why Morgan Stanley’s Revision Matters for Capital Planning

For sophisticated investors, Morgan Stanley’s analysis highlights a critical constraint beneath the AI infrastructure cycle: computing capacity cannot scale independently of power availability. The bank’s higher shortfall forecast suggests that electricity infrastructure, onsite generation and access to existing power assets are becoming increasingly important variables in the deployment timetable for AI capacity.

The next phase of Morgan Stanley’s analysis will therefore depend on how quickly accelerated generation, grid access and alternative power arrangements can close the projected gap. For global capital allocators, the bank’s revised framework places power availability alongside chips, data centers and networking as a central consideration in the broader AI infrastructure buildout.

For a confidential discussion regarding your cross-border banking structure, technology infrastructure exposure or international wealth strategy, contact our senior advisory team.