Morgan Stanley Says AI Is About to Have Its Biggest Breakthrough Yet. There Is One Problem: Not Enough Electricity.
The bank predicts a massive AI capability leap in 2026. But the US faces a 9 to 18 gigawatt power shortfall that could derail everything.
The AI Post newsroom — delivering AI news at the speed of intelligence.
Morgan Stanley just published a report that should make every AI investor both excited and terrified at the same time.
The excitement: the bank's analysts predict that the rapid scaling of compute infrastructure at major AI labs will trigger a transformative leap in AI capabilities this year. OpenAI's latest model reportedly scored 83% on the GDPVal benchmark, which measures expert-level performance on complex economic tasks. Scaling laws, they say, are still holding. More compute equals smarter models. And compute is growing exponentially.
The terror: there is not enough electricity to power it.
Morgan Stanley's "Intelligence Factory" model predicts the United States will face a power shortfall of 9 to 18 gigawatts by 2028. That is a 12% to 25% gap between what AI data centers need and what the grid can actually deliver. For context, 18 gigawatts is roughly the output of 18 nuclear power plants. We do not have 18 spare nuclear power plants.
The solutions being explored are creative, desperate, or both. Companies are converting old Bitcoin mining facilities into AI compute hubs. Others are installing natural gas turbines on-site to generate their own electricity. Fuel cells are being deployed as backup. Microsoft tried putting data centers underwater and abandoned the idea. SpaceX wants to put them in orbit. None of these solve the core problem: the AI industry is growing faster than the power grid can keep up.
The economics are staggering. Morgan Stanley describes a "15-15-15" model emerging in AI infrastructure: 15-year lease agreements, 15% investment yields, and around 15% annual growth in demand. Data center debt issuance hit $165 billion in 2025. The bank expects that to more than double to over $400 billion this year. This is not a bubble. It is an infrastructure supercycle with a very real constraint at its center.
Here is the uncomfortable truth the AI industry does not want to talk about: the breakthrough everyone is racing toward might arrive on schedule, but the infrastructure to support it might not. The models will get smarter. The question is whether we can keep the lights on long enough to run them.
The companies that figure out energy first will win the AI race. Not the ones with the best models. The ones with the most watts.