Editor’s correction, September 9, 2026: this article was first published on March 14, 2026. We have replaced unsupported claims about an imminent AI breakthrough, job replacement and investment returns with a more careful account of the energy outlook. Forecasts below are identified by their source and date.
AI lives in software, but it still needs somewhere to plug in. That makes electricity an important part of the story whenever a technology forecast promises much larger models or much wider adoption. It does not, however, make an energy forecast proof that a particular breakthrough will arrive on schedule.
What Morgan Stanley actually described
In its February 27, 2026 energy outlook, Morgan Stanley described developers’ concerns about power constraints in 2027 and 2028. Its account connected those concerns to grid investment, equipment supply and the difficulty of connecting large new facilities. It also discussed on-site and hybrid power arrangements, including gas, renewables, storage and nuclear options.
That is an account of infrastructure pressures and possible responses. It is not a demonstration that every AI developer will run out of electricity, or that a specific model will suddenly replace most human work. The article also reflects an investment perspective: possible opportunities for equipment suppliers and energy providers should be read as expectations, with execution risk, rather than guaranteed returns.
A scenario is not a meter reading
The International Energy Agency offers a useful separate reference. In the Base Case of its 2025 Energy and AI report, worldwide data-centre electricity consumption was projected to reach about 945 terawatt-hours in 2030, more than double its 2024 level and just under 3% of global electricity demand. These are projections from that report, not measured 2030 consumption or a claim to be the latest forecast.
The category also matters: data centres support more than generative AI. A number for the whole sector should not be relabelled as electricity used exclusively by chatbots. The report examines uncertainty around deployment and efficiency, which is another reason to preserve the scenario label when repeating its headline figures.
Three questions to ask of an AI headline
First, what is being measured? A model’s result on a particular evaluation, the capacity of a planned facility and the electricity consumed over a year answer different questions. Putting them in one dramatic paragraph does not establish a causal link between them.
Second, what is the time frame? An announced project is not the same thing as an operating facility. A forecast about 2028 is not evidence of a shortage today. Keeping publication dates and forecast dates visible helps prevent an old prediction from masquerading as a new observation.
Third, what would change the conclusion? A useful outlook should leave room for slower adoption, efficiency improvements, delays or different technology choices. If a headline only allows for one inevitable outcome, it is worth returning to the underlying source before drawing a personal or business conclusion.
Keep the excitement, keep the distinctions
The interesting story is the connection between digital ambitions and physical infrastructure. We can follow that story without treating uncertain productivity gains as settled facts or turning a discussion of power supply into a stock tip. Our earlier version blurred those distinctions; this correction restores them.
🐾 Curious about what comes next? Explore more of our Technology stories.
