← All Insights

2026-10-23 · 阅读中文版 → · Leer este artículo en español →

Why Power — Not Land — Is Now the Binding Constraint on AI Data Centers

For most of commercial real estate history, the scarce input for a large project was the site itself: the right parcel, in the right location, with the right zoning. Data center development has inverted that order. Across most of the United States, suitable land is widely available. What is scarce — and what now determines whether a project happens at all — is electric power a utility can actually deliver, at the required scale, on a timeline a developer can finance. That is what the major energy and market datasets show when read together.

Demand has stepped up faster than the grid's planning cycle

Start with the scale of demand. The International Energy Agency's Energy and AI report (April 2025) estimates that data centers consumed about 415 TWh of electricity globally in 2024, roughly 1.5% of world electricity demand, and projects consumption to more than double to around 945 TWh by 2030 in its base case. The IEA identifies AI as the most significant driver: electricity demand from AI-optimised data centres is projected to more than quadruple by 2030, and the United States accounts for the largest national increase, with U.S. data center consumption rising by up to about 240 TWh (roughly +130% versus 2024). In the IEA's assessment, U.S. data center power consumption is on course to account for almost half of the growth in total U.S. electricity demand between now and 2030.

The U.S. national laboratory data tells the same story from the bottom up. Lawrence Berkeley National Laboratory's 2024 United States Data Center Energy Usage Report (December 2024) found that U.S. data center electricity use climbed from 58 TWh in 2014 to 176 TWh in 2023 — about 4.4% of total U.S. electricity — and estimated it could reach between 325 and 580 TWh by 2028, or 6.7% to 12% of national consumption. The report notes that data center power demand more than doubled between 2017 and 2023, largely due to the growth in AI servers, and that load growth had already tripled over the preceding decade.

EPRI, the electric power industry's research institute, has repeatedly revised its outlook upward. Its 2024 Powering Intelligence study modeled data centers reaching 4.6% to 9.1% of U.S. electricity generation by 2030; its February 2026 update raised that range to 9% to 17%, with estimates roughly 60% higher, driven by the surge of announced and under-construction projects. National averages also understate the problem, because demand is geographically concentrated: fifteen states account for about 80% of the national data center load, and in Virginia — the largest market — data centers already consumed around a quarter of the state's electricity in 2023 (EPRI, 2024).

A single facility explains the local impact. EPRI notes it is not unusual for new data centers to be built with capacities from 100 to 1,000 MW — roughly the load of 80,000 to 800,000 homes. A load of that size cannot plug into a distribution network designed for a warehouse park; it requires transmission-level planning, substation capacity, and generation behind it.

Land is rarely the binding constraint

Against that demand picture, land supply is not the limiting factor in most regions. The United States has extensive industrially zoned land, former manufacturing sites, and greenfield parcels within reach of transmission infrastructure. CBRE's reporting makes the point clearly: in 2025, growth spread toward markets such as Nevada, Pennsylvania and Michigan, which CBRE describes as attractive because of abundant land, more flexible permitting and potentially easier access to power. Land can be found in many places; the differentiator is access to power.

This is why experienced developers now describe their process as "power-first." CBRE's Pat Lynch, summarising its North America Data Center Trends reporting for 2025, put it this way: demand is outpacing supply, and "power and supply chain shortages are reinforcing a power-first approach that prioritizes sites with the fastest path to power."

Power arrives on a grid timetable, not a construction timetable

The core problem is timing. A modern data center building can be designed and constructed on a schedule measured in a small number of years. The grid reinforcements behind it — interconnection studies, transmission upgrades, substations, transformers, and in many cases new generation — run on longer cycles, and the queue data shows those cycles lengthening.

Berkeley Lab's Queued Up: 2024 Edition found about 2.6 TW of planned generation and storage capacity active in U.S. interconnection queues at the end of 2023 — roughly twice the country's existing installed generating capacity, and up 27% year over year. Two findings matter for anyone evaluating a site. First, only a small share of queued capacity had cleared the final step: projects totaling 311 GW, about 12% of queued capacity, had executed interconnection agreements. Second, completion rates are historically low: only about 14% of the capacity that entered queues between 2000 and 2018 had been built by the end of 2023. Wait times have stretched accordingly — the typical project built in 2023 took nearly five years from interconnection request to commercial operations, compared with about three years in 2015 and under two years in 2008.

Two cautions apply to queue data. The Queued Up figures describe generation and storage projects, not data center loads; large-load service requests run through separate utility processes. And a queue position is an application, not a commitment — evidence that capacity has been requested, not that it has been granted, funded, or scheduled.

The market consequences are visible in CBRE's numbers. Across the eight primary North American markets, vacancy fell to a historic low of 1.4% at the end of 2025, even though total capacity grew 36% year over year to 9,432 MW and users absorbed 2,497.6 MW during the year. Yet capacity under construction declined — to 5,994.4 MW from 6,350.1 MW, the first decline since 2020 — which CBRE attributes to extended timelines tied to permitting, zoning approvals and sourcing adequate power. In CBRE's H1 2025 reporting, 74.3% of capacity under construction was already preleased, as cloud and AI users moved to secure future power years in advance.

Read together: buildings can be delivered faster than the power behind them, tenants know it, and they are contracting for capacity long before it exists as an operating facility.

How to read a capacity claim

This is where discipline matters, because the industry quotes very large megawatt numbers with very different levels of substance behind them. We classify every capacity figure into one of four states, and never mix them:

A site marketed as a "500 MW opportunity" is usually describing planned or applied capacity. The financed, deliverable reality may be a fraction of that, years later. None of this makes the site bad — early-stage capacity is where development value is created — but the four states carry completely different risk, and any serious evaluation should state, for each megawatt figure, which state it is in and what document proves it.

What this means for site selection

If power is the binding constraint, then site screening has to be reordered. The traditional sequence — find land, check zoning, then ask the utility what it can deliver — burns months and diligence budget before reaching the question that decides the project. The working sequence now is:

  1. Define the load precisely: target MW, phasing, and the date capacity is actually needed.
  2. Establish the utility service territory and the evidence for deliverable capacity at or near the site, classified by the four states above.
  3. Only then spend on the remaining diligence — title, environmental, geotechnical, full permitting analysis.

Land still matters: zoning, water, fiber, hazards and community acceptance can each stop a project. But they are checks that follow the power question, not substitutes for it. In today's market, a site is an opportunity because of the power that can be proven to reach it — and the date that power can be proven to arrive.

Sources

  1. International Energy Agency (IEA), Energy and AI, April 2025. Global data center electricity consumption of about 415 TWh in 2024 (~1.5% of global demand), projected to more than double to around 945 TWh by 2030 (base case); AI-optimised data center demand projected to more than quadruple; U.S. increase of up to ~240 TWh (+130%); U.S. data centers on course to account for almost half of U.S. electricity demand growth to 2030. Accessed via IEA report summaries and Data Center Dynamics coverage, 2026-10-23. https://www.datacenterdynamics.com/en/news/iea-data-center-energy-consumption-set-to-double-by-2030-to-945twh/ — Evidence grade: A (primary institutional source; figures as published by the IEA).
  2. Lawrence Berkeley National Laboratory (LBNL), 2024 United States Data Center Energy Usage Report, December 2024. U.S. data center consumption 58 TWh (2014) → 176 TWh (2023, 4.4% of U.S. electricity), estimated 325–580 TWh by 2028 (6.7%–12%); demand more than doubled 2017–2023, largely due to AI servers. Berkeley Lab News Center, January 15, 2025. https://newscenter.lbl.gov/2025/01/15/berkeley-lab-report-evaluates-increase-in-electricity-demand-from-data-centers/ — Evidence grade: A.
  3. EPRI, Powering Intelligence: Analyzing Artificial Intelligence and Data Center Energy Consumption, 2024. Scenarios of data centers reaching 4.6%–9.1% of U.S. electricity generation by 2030; fifteen states account for ~80% of national data center load; Virginia data centers ~25% of state load in 2023; new facilities commonly 100–1,000 MW. Summary: American Public Power Association, 2024. https://www.publicpower.org/periodical/article/epri-report-examines-power-demand-data-centers-artificial-intelligence — Evidence grade: B.
  4. EPRI, Powering Intelligence 2026: Updated Scenarios of U.S. Data Center Electricity Use and Power Strategies, February 25–26, 2026. Updated range of 9%–17% of U.S. electricity by 2030, ~60% higher than 2024 estimates. EPRI press release via GlobeNewswire, February 26, 2026. https://pnr-files.pro1.gus.wdc.dianum.io/globenewswire/articles/3245491/en/epri-data-centers-could-consume-up-to-17-of-u-s.pdf — Evidence grade: B.
  5. Lawrence Berkeley National Laboratory, Queued Up: 2024 Edition — Characteristics of Power Plants Seeking Transmission Interconnection, April 2024. ~2.6 TW active in queues at end of 2023; 311 GW (~12%) with executed interconnection agreements; ~14% of 2000–2018 queued capacity built; typical project built in 2023 took nearly five years from request to operations. Reported by Utility Dive, April 11, 2024. https://www.utilitydive.com/news/grid-interconnection-queue-berkeley-lab-lbnl/712926/ — Evidence grade: A/B (primary research, accessed via trade-press report of the release).
  6. CBRE, North America Data Center Trends (H2 2025), press release "Fast-Growing North American Data Center Market Set Records in 2025," 2026. Vacancy 1.4% (historic low); total primary-market capacity 9,432 MW (+36%); net absorption 2,497.6 MW in 2025; capacity under construction 5,994.4 MW, down from 6,350.1 MW, first decline since 2020, attributed to permitting, zoning and power-sourcing timelines; "power-first" characterisation. https://www.cbre.com/press-releases/fast-growing-north-american-data-center-market-set-records-in-2025 — Evidence grade: B.
  7. CBRE, North America Data Center Trends H1 2025, reported by Data Center Dynamics, September 2025. Vacancy 1.6% in H1 2025; 74.3% of capacity under construction preleased. https://www.datacenterdynamics.com/en/news/vacancy-rates-in-north-american-data-center-markets-hit-record-low-cbre/ — Evidence grade: B/C.