Power now decides most data-centre sites. Land is cheaper to fix than power. A plot with a firm power date beats a bigger plot without one.
This article explains why, what the best public research says, and what to ask a utility before you pay for land.
Why has power become the first filter?
Demand has grown faster than the grid. Three public sources show the scale.
- United States. The Lawrence Berkeley National Laboratory report, released by the US Department of Energy in December 2024, found that data centres used about 4.4% of US electricity in 2023. It projects 6.7% to 12% by 2028. Usage rose from 58 TWh in 2014 to 176 TWh in 2023, and the report gives 325 to 580 TWh for 2028.
- World. The International Energy Agency says data centres used about 415 TWh in 2024, around 1.5% of world electricity. Its base case is about 945 TWh by 2030.
- Concentration. The IEA reports that the United States was 45% of the 2024 total, and that nearly half of US data-centre capacity sits in five regional clusters. It also finds that about 20% of planned data-centre projects could be at risk of delay, and that grid connection queues "are long and complex".
The 2028 and 2030 numbers are scenarios. They are not forecasts you can bank on. The spread between 6.7% and 12% is the point: nobody knows yet.
What does a crowded queue do to everyone?
It pushes cost and risk onto the whole region, not only the data-centre owner.
PJM is the grid operator for a large part of the eastern United States. Its independent market monitor told PJM members in August 2026 that existing and forecast data-centre load added $29.4 billion to capacity-market revenue across the last four auctions. It also said data-centre load made up 9% of PJM's wholesale power price through July 2026, and warned that the total "will continue to grow until the issues associated with the addition of large data center loads are addressed." (Reported by Utility Dive, 27 August 2026.)
This matters to a developer for a simple reason. Regulators and utilities react to cost pressure. Expect more study fees, more proof of seriousness, and tighter rules on who can join a queue. A site plan that assumes today's rules may meet different ones when the application is filed.
What do the wait figures say?
Published wait times differ a lot. Advisory firms report average grid-connection waits above four years in major markets, ranges of three to seven years, and more than six years in parts of PJM. Heavy equipment adds its own delay: transformers, switchgear and large UPS systems are reported to take about one and a half to four years to deliver, depending on type and size.
Treat these as a signal, not a schedule. They come from advisory and vendor writing, not from a utility filing. Two sites in the same state can have very different dates. The only number that matters is the one the serving utility gives you in writing for your load.
Ten questions to ask the utility before you buy land
- Who serves this parcel, and is the nearest substation owned by the same company?
- What is the voltage at the nearest line, and at what distance?
- How much load is available today at that substation, in MW, without new build?
- What must be built (line, substation, transformer) to serve your full load?
- Who pays for it, and is any part refundable?
- What is the study process, the fee, and the date the study would finish?
- Is there a queue, and where would your request sit in it?
- Can the load ramp in phases, for example 20 MW first, then more?
- What are the equipment lead times for the transformers and switchgear your design needs?
- What are the conditions if you bring your own generation, and how does that change the interconnection study?
Get the answers in writing. A phone call is not a commitment.
What can a developer do before the answers arrive?
- Screen power first, land second. Rank candidate sites by the utility's answer, then by price and zoning.
- Pay attention to existing clusters. The IEA finds half of US projects under development sit in existing large clusters. That helps fibre and skills. It can also stack many projects on one substation.
- Keep a second route. On-site generation and storage are being discussed by the US Department of Energy as a way for data centres to become "a grid asset rather than a burden". They add cost and permitting. Price them early, not late.
- Do not count the queue as capacity. A request in a queue is a hope. Energisation is the event.
What this does not tell you
- It does not give a wait time for your site. Only the utility can.
- The 2028 and 2030 projections are scenarios, and some of the queue in any market is speculative.
- India has its own grid structure and approval path. Read our article on India's data-centre capacity by city before you apply these questions there.
Sources
- Lawrence Berkeley National Laboratory, *2024 United States Data Center Energy Usage Report* (December 2024), summarised by the US Department of Energy: energy.gov.
- International Energy Agency, *Energy and AI*, executive summary: iea.org.
- PJM independent market monitor presentation, reported by Utility Dive, 27 August 2026: utilitydive.com.
- Queue-wait and equipment lead-time figures: advisory and vendor writing, for example Landgate and Build Central. Not primary filings. Checked 21 October 2026.
Keystone AEC (keystoneaec.com) is an AI platform for land feasibility and master planning in India and the USA. Bring a site and a load, and the useful first conversation is about which of these ten answers you still lack.