Morgan Stanley Raises US Data Center Power Shortfall Estimate to 57 GW for 2026–2028
DataNews Editorial Team
Morgan Stanley has raised its estimate of the gross US data-center power shortfall for 2026–2028 to 57 GW, from 38 GW previously. Its probability-weighted central estimate for the net shortfall is 33 GW, reflecting higher expected demand from increasingly dense AI rack architectures.
- Morgan Stanley lifted its gross US data-center power-shortfall estimate for 2026–2028 to 57 GW, from 38 GW previously.
- Projected aggregate US data-center electricity demand is 97 GW for the period.
- The bank’s probability-weighted central case puts the net shortfall at 33 GW after fast-track supply measures.
- Higher-density AI rack architectures, including NVL72 systems, underpin the demand revision.
- Behind-the-meter generation, fuel cells and siting near operating nuclear plants are modeled as potential deficit-reduction measures.
Morgan Stanley has materially increased its outlook for the US data-center power deficit in 2026–2028, according to a report published on 21 September and cited by BigGo Finance. The bank now forecasts a gross shortfall of 57 GW, up from its prior 38 GW estimate.
The revised balance begins with projected aggregate US data-center electricity demand of 97 GW for the 2026–2028 period. Morgan Stanley credits 21 GW of capacity associated with data centers under construction and 19 GW of available grid capacity; the remaining gross gap is 57 GW.
After applying probability-weighted fast-track supply measures, Morgan Stanley’s central estimate for the net shortfall is 33 GW. Its scenario analysis spans a 42 GW net deficit in the low case, 33 GW in the central case and no net deficit in the high case. These are forecasts rather than commissioned or guaranteed power supplies.
The revision is tied to a shift in AI infrastructure toward NVL72 rack systems, moving from eight-GPU configurations to racks with 72 GPUs and associated equipment. Morgan Stanley said this architecture raises both power and cooling requirements. It increased its US power-demand forecasts by 38% for 2027 and 69% for 2028 versus its previous model.
The supply-side mitigation case includes a probability-weighted 19 GW contribution from new behind-the-meter gas turbines and reciprocating generators, 6 GW from Bloom Energy fuel cells, and 3 GW from locating data centers near operating nuclear power plants. The bank estimates the value of securing power one year earlier at about $4.5/W.
For developers, colocation operators and hyperscale customers, the implication is that deliverable power and the cooling infrastructure capable of supporting high-density racks could become more decisive than site area or announced capacity. The analysis also points to potential growth in behind-the-meter generation transactions, particularly in West Texas, although the extent of this activity will depend on whether the modeled supply measures can be executed on schedule.
The assessment reinforces power delivery and cooling capacity—not simply server supply or data-center real estate—as the primary constraints on US AI infrastructure deployment. If the forecast materialises, operators and developers with secured interconnection capacity, available power, or the ability to deploy local generation quickly could hold an advantage over projects with only nominal future capacity.
- — / September 22, 2026 · JapaneseSource →
This article was prepared by the DataNews editorial team based on the sources listed above.