More than 2,600 gigawatts of proposed generation and storage now sit in U.S. interconnection queues — over double the country's entire existing operational grid capacity. Only 13% of requests filed between 2000 and 2019 ever reached operation. In 2026, the hardest part of building a data center isn't the building. It's getting the utility to actually turn the power on.
Every conversation about the AI infrastructure buildout eventually arrives at the same uncomfortable fact: it isn't chips, land, or capital holding back new data center capacity in 2026. DigitalBridge's CEO put it in four words: "power is really the constraining factor." Here's exactly how big that constraint is, why it's structural rather than temporary, and what operators are doing to route around it.
Data center power capacity, by the numbers
That last figure explains the urgency behind everything else in this article. Accelerating a campus launch by even two years ahead of a grid connection can mean tens of billions of dollars in additional revenue — which is exactly why operators are now willing to pay a premium, take on construction risk, and pursue power strategies that would have seemed extreme just three years ago.
The interconnection queue crisis
A case study in overload: ERCOT
Texas is the sharpest illustration of the gap between announced ambition and physical reality anywhere in the country.
Of the 16 GW of new data center power capacity originally targeted for 2026 nationally, only 5 GW had entered active construction as of early 2026, with industry analysts projecting 30–50% of the remainder slipping to 2027 or later. Northern Virginia, the world's largest data center market, illustrates the resulting scarcity directly: colocation vacancy sits at just 0.72%, with 87% of all inventory scheduled for 2025–2026 delivery already preleased before it's even built.
Why "getting power" takes years, not months
Rack density is compounding the problem
It isn't just that more data centers are being built — each one is drawing dramatically more power per square foot than its predecessor. That density shift means the same physical footprint that once needed a single substation upgrade now needs several, compressing years of expected grid-capacity headroom into a fraction of the time planners originally modeled.
The bring-your-own-power shift
Faced with multi-year grid delays, a growing share of operators are simply generating their own power on-site rather than waiting — gas turbines, and increasingly small modular reactors, deployed specifically to bypass the interconnection queue entirely. This is a genuinely new strategic posture for the industry: data center developers historically thought of themselves as power consumers, not power producers, and that distinction is eroding fast.
What regulators are actually doing about it
- FERC Order 2023 replaced the old serial, first-come-first-served interconnection study process with cluster-based studies, and added stricter site-control requirements, higher deposits, and withdrawal penalties specifically designed to discourage the speculative multi-site filing behavior described in the ERCOT section above.
- The gap Order 2023 doesn't close: it primarily addresses generation interconnection, not large-load interconnection — meaning the specific queue category data centers fall into remains comparatively under-reformed.
- DOE's October 2025 directive proposed federal jurisdiction over new electricity loads above 20 megawatts, an attempt to standardize and potentially accelerate a process that currently varies enormously by state and utility.
- Texas SB6 gives grid operators explicit authority to curtail large-load power during emergencies — a policy response that introduces real uptime risk for compute-critical AI workloads, even after a facility successfully secures a grid connection.
What this means for site selection and rents
Power availability has become the primary variable in data center site selection, ahead of land cost, tax incentives, or even fiber connectivity in many cases. Markets that can offer a credible, fast path to energization command a premium; markets where the queue stretches past five years are being deprioritized regardless of other advantages. This dynamic connects directly to the extreme scarcity pricing already visible in constrained markets like Northern Virginia, where sub-1% vacancy and near-total preleasing of unbuilt inventory reflect power scarcity as much as demand strength.
Choosing a power strategy for your project
For the specific SMR pathway several hyperscalers are already pursuing to solve this exact bottleneck, see our SMR nuclear power for data centers deep dive, and for how rising rack density is separately reshaping cooling architecture, see our liquid vs. air cooling cost comparison.
Frequently asked questions
Grid interconnection queues have grown faster than utilities can process them — over 2,600 GW of generation and storage is currently stuck in U.S. queues, more than double the country's existing installed capacity, with only about 13% of older requests ever reaching operational status. Equipment shortages (128-week transformer lead times) compound the delay even after grid approval is granted.
It varies enormously by market, but multi-year delays are now typical — commonly 4 years or more from interconnection request to operation, and in some major hubs like Northern Virginia, Dublin, Singapore, and Amsterdam, 4 to 7 years. In the most constrained cases, grid-only strategies can take up to 10 years, and one utility reportedly quoted 12 years just to evaluate a single request.
BYOP means a data center generates its own power on-site (gas turbines, or increasingly small modular reactors) rather than waiting for a grid connection. Roughly 30% of planned U.S. data center capacity is shifting toward this model, driven by the fact that a two-year head start on launch can mean tens of billions in additional revenue at $10–12M per megawatt annually.
The headline number (2,600+ GW) is real, but it's inflated by speculative filings — roughly 70% of Texas's queue hadn't reached the first paid engineering checkpoint that separates serious projects from speculative ones, since joining a queue has historically been close to free. FERC's Order 2023 specifically targets this behavior with stricter site-control and deposit requirements.
Our methodology
Every capacity, timeline, and queue figure in this article is sourced from a named grid operator (ERCOT), federal agency (DOE, FERC), research institution (Lawrence Berkeley National Laboratory), or named industry survey (Bloom Energy's 2026 Data Center Power Report). Where sources gave differing totals for the same metric (interconnection queue size cited as 1,500 GW, 2,300 GW, and 2,600 GW across different sources and dates), we attributed each figure to its specific source and date rather than picking one to imply false precision.
- Interconnection queue figures change month to month as new projects file and speculative ones are withdrawn under stricter FERC rules; treat these as directional, not exact, at any given moment.
- ERCOT-specific figures reflect its March 2026 published update and its most recent large-load queue disclosure; other regions (PJM, MISO) face similar but not identical dynamics not covered in equivalent depth here.
- We did not include vendor-specific marketing claims about power solutions without independent corroboration from a named grid operator, regulator, or research source.
- This article is reviewed periodically as new queue data, regulatory actions, and hyperscaler power deals are announced.
Sources
Data compiled from the following named sources (accessed August 2026):
- Data Center Knowledge — "Gridlocked: Power Constraints Shape the Future of Data Centers," March 2026
- EnkiAI — "AI & Data Center Energy 2026, 2,600 GW Queue & PJM Plan," June 2026
- Bloom Energy — "2026 Data Center Power Report," January 2026 survey of hyperscaler energy strategy leaders
- Global Data Center Hub — "ERCOT Data Center Queue and Energized Capacity," March 2026 ERCOT update
- mGrid — "Data Center Grid Delays Put 50% of 2026 AI Capacity at Risk," January 2026
- TechPlusTrends — "AI Data Center Power Requirements 2026: The Grid-to-Chip Guide," citing Lawrence Berkeley National Laboratory interconnection queue data
- Hanwha Data Centers — "Data Center Grid Limitations: The Power Bottleneck," February 2026, including FERC Order 2023 and DOE October 2025 directive details
- iRecruit — "Data Center Power & Energy News 2026: Weekly Project Updates"
- U.S. Department of Energy — October 2025 directive on federal jurisdiction over large loads
