Short answer
Grids can carry the load in most places over time, but not at the speed AI companies want, so connection queues, on-site gas plants and fights over cost are now the main bottlenecks. Who pays is being decided right now by regulators, from the largest US grid to Hydro-Québec, and the answers differ from place to place.
Follow-up question Will AI data centers raise your electricity bill?
A large AI campus wants more electricity than some cities, and it wants it in one or two years. Building a new power line can take far longer: a University of Alberta engineering professor told CBC that transmission alone can take 10 years. That mismatch in timing, more than any shortage of electricity worldwide, is the heart of the problem.
The global share is still modest. The International Energy Agency (IEA) expects data centers to use about 3% of the world's electricity in 2030. The strain is local, because these sites cluster. In the United States, the Department of Energy reported that data centers used about 4.4% of electricity in 2023 and could reach 6.7% to 12% by 2028.
Why connecting takes so long
Grid operators must check that a new large customer will not overload the system, then build any lines and substations it needs. Requests have overwhelmed that process. New York cited nearly 12 gigawatts of data center requests in its grid queue, with more than 8 entering in 2025 alone. In Alberta, 37 projects were asking for 19.4 gigawatts in November 2025, after the first allocation was used up.
The IEA notes that not all of these projects will be built. Grid planners have to judge which ones are real, and the PJM case below shows how much money can rest on that forecast.
How builders get around the wait
They build their own power. The IEA says US developers, slowed by grid connections, are pushing ahead with on-site gas plants, and satellite tracking shows about a fifth of those projects have started land clearing or construction. On-site power is not simple. The IEA says reliable supply requires building 30% to 70% more generation than the load, and a shortage of gas turbines may erase the time saved.
They pay a utility to build for them. Meta's Louisiana campus is the largest example. Entergy Louisiana announced an agreement in March 2026 that includes seven new gas plants totaling more than 5,200 megawatts. Fortune reports that, with three approved earlier, Meta would pay for 10 plants totaling 7.5 gigawatts at nearly $11 billion, more than 30% of the state's grid capacity. Meta also agreed to help fund up to 2,500 megawatts of renewable power.
They add batteries. AI work causes large, fast swings in demand. The IEA expects 20 to 25 gigawatts of battery storage in data centers by 2030, which could also help the wider grid if the incentives are right.
Who pays: three real cases
PJM, the largest US grid operator. PJM runs a capacity auction, in which power plants are paid to be available in future years, and everyone's bills cover the cost. Its independent market monitor found that data center load drove about $6.5 billion of the December 2025 auction's $16.4 billion cost, a 40% share. About $6.2 billion of that came from data centers not yet built. PJM also fell 6,516.6 megawatts short of its reliability target. In October 2026, federal regulators rejected most of PJM's plan for connecting large loads, saying key parts intrude on state authority over retail rates.
Louisiana. Entergy says Meta pays its full cost of service and expects about $2 billion in customer savings over 20 years, on top of $650 million announced earlier. These are the utility's projections. Critics quoted by Fortune worry that customers could carry the plants after the 15-year contracts end, if Meta no longer needs that much power.
Québec. Hydro-Québec has asked the Régie de l'énergie, the provincial energy regulator, to approve a new rate for data centers above 5 megawatts, averaging 13 cents per kilowatt-hour. At the hearing, the utility put today's average at 6.82 cents and its energy cost at about 12 cents, and said the period of surplus power is over. It said the rate is not a moratorium. Google has asked the regulator to reject the proposal, while environmental groups are also intervening. The hearing opened in early October, and the Régie's ruling is the next thing to watch.
A worked example: one Québec bill
Take a data center drawing a steady 100 megawatts, which is half of the province's current peak for the sector.
| Item | Calculation | Result |
|---|---|---|
| Energy in a year | 100 MW × 8,760 hours | 876,000 MWh |
| Cost at 6.82 ¢/kWh | 876 million kWh × $0.0682 | about $60 million |
| Cost at 13 ¢/kWh | 876 million kWh × $0.13 | about $114 million |
Illustrative: real bills depend on the rate structure, and existing customers would get a five-year transition. A difference of $54 million a year is large, but small next to the billions spent on chips for a site that size.
How other provinces are rationing
Provinces with limited supply now choose who connects.
| Province | Current approach |
|---|---|
| Québec | Ministerial authorization required to supply 5 MW or more |
| British Columbia | 100 MW conventional and 300 MW AI over two years, 145 MW maximum per request, through a competitive call |
| Alberta | 1,200 MW interim limit fully allocated; a new regulation favors projects bringing their own generation |
| Ontario | Consulted on government approval for large connections and a separate rate class |
Try it: Look up your electricity provider's latest rate filing or news release and search for "data centre" or "large load". Note whether new large customers pay their own connection costs.
Will AI data centers raise your electricity bill?
It depends on where you live and on rules still being written. In a market like PJM, the market monitor's numbers show that data center demand has already raised capacity costs, which flow to customers. Hydro-Québec cites Bloomberg figures showing electricity bills more than doubling over five years in US areas with strong data center growth, and that comparison is the utility's own framing for its rate request.
The opposite argument also exists: a large customer that pays its full cost can spread fixed grid costs over more electricity sold, which is the basis of Entergy's projected savings. Whether that happens depends on contract terms, and on what happens if the customer leaves. The practical questions to ask of any project are who pays for new lines and plants, and who is left paying if demand falls short.
Sources
- Key Questions on Energy and AI: Executive summary — International Energy Agency, April 16, 2026
- Data centers were 40% of PJM capacity costs in last auction: market monitor — Utility Dive, January 6, 2026
- A new rate for data centres and a rate adjustment for blockchains — Hydro-Québec, February 19, 2026
From the Guide to Artificial Intelligence, by Robert Scales, published by Holistic Hybrid Inc. Share and adapt it under CC BY-NC-SA 4.0: credit the guide, link to https://guidetoartificialintelligence.com/en/learn/ai-data-center-power-grid/, no commercial use without permission, and keep the same licence.