There's a real power shortage for AI data centers and this chart puts an actual number on it (Save this).
Milk Road AI Twitter · Milk Road AI (@MilkRoadAI) · 2026-08-03
Milk Road AI summarizes a Morgan Stanley analysis projecting a 38GW US power shortfall for AI data centers through 2028 and recommends investment positions in natural gas turbine makers, fuel cell companies, nuclear operators, and converted Bitcoin mining sites.
Appears in
Extraction
Topics: ai-data-center-powerenergy-infrastructureai-investment-thesis
Claims
- Morgan Stanley projects US data centers will need 68 gigawatts of power between 2026 and 2028.
- After accounting for data centers under construction and contracted grid capacity, a raw shortfall of 38GW remains.
- Natural gas turbines are the largest addressable lever at 15-20GW with 90% probability of success.
- Even after probability-weighting all near-term solutions, Morgan Stanley projects a net shortfall of 1 to 11GW through 2028.
- Converting Bitcoin mining sites with existing power infrastructure to AI compute is treated as a meaningful part of the solution set.
Key quotes
Power is the actual constraint on how fast the AI buildout can scale right now because you can have all the GPUs in the world sitting in a warehouse, but if there's no electricity to run them, they're stranded capital.
Morgan Stanley still lands on a net shortfall of 1 to 11GW through 2028, with their base case being a narrow but real gap.
Natural gas turbines are the biggest lever at a 15-20GW range with 90% probability of success.