Morgan Stanley estimates that AI data centers will face a 38-gigawatt power shortfall through 2028. The investment bank's projection underscores a growing strain on electricity grids as AI computing demand ramps up.
Why the gap is so wide
The shortfall is not a single year's problem. It spans the entire period through 2028, meaning the gap builds up each year as new data centers come online. 38 gigawatts is a massive amount of electricity — more than many countries generate in total. But the bank's analysts didn't specify which regions or grid systems will be hit hardest, only that the overall need for power will outrun supply.
AI data centers are energy-hungry operations. They run thousands of processors, need constant cooling, and often operate around the clock. That baseline demand is what drives the forecast. The estimate from Morgan Stanley is a forward-looking warning, not a report on current shortages.
What this means for the AI buildout
The immediate implication is that building new AI infrastructure is no longer just about finding land, chips, or water — it's about securing enough electricity. If the shortfall holds, some planned data centers could face delays or have to wait for grid upgrades.
That doesn't mean AI development will stop. It does mean the pace of expansion could slow unless power generation and transmission catch up. The bank's estimate doesn't say who will solve the problem or how. It simply lays out the scale of the challenge.
For utilities and grid operators, the takeaway is more direct: they need to plan for a world where AI centers consume a lot more power than they do today. For companies building those centers, the forecast is a reminder that energy is a constraint on growth.
The numbers cover the period through 2028, so the clock is already ticking. The next few years will show whether new generation and grid upgrades can close the gap — or whether the shortfall becomes a hard limit on AI's expansion.
