Beyond the Announcement: Assessing the Deliverability of the U.S. Data Center Pipeline

Executive Summary

The core finding is not that hundreds of gigawatts are already cancelled. The evidence establishes a more precise conclusion: a very large share of the U.S. data-center pipeline remains unbuilt, and projects targeted for near-term delivery are materially less advanced than headline announcements indicate. Currence tracks 190 GW across 777 large data-center and AI-factory projects announced since 2024. In its February 2026 outlook, only approximately 5 GW of the 16 GW targeted for 2026 is under construction, leaving approximately 11 GW in the announced stage without visible construction progress at that point.

The two national datasets are not additive. They use different coverage, reporting dates, inclusion criteria, project definitions, and status classifications. Currence/Sightline therefore serves as the primary source for the 2026 execution-risk analysis, while DCMap serves as the primary source for the state-by-state pipeline analysis. This approach maintains a clear audit trail and prevents the creation of a misleading combined national capacity figure.

What Counts as Announced, Deliverable, and at Risk?

A data-center project passes through multiple execution stages between initial announcement and commercial operation. A public announcement establishes planned development, but it does not establish secured land, power, permits, financing, customer commitments, or construction. A utility large-load or interconnection request establishes demand for grid capacity, but it does not establish a financeable or executable data-center project. A developer can also secure a site and construction agreement while generation or transmission capacity required for energization remains unresolved.

The report therefore separates projects according to the strength of their execution evidence. This classification prevents announced capacity, grid requests, and projects with confirmed construction activity from being treated as equivalent.

This distinction is critical because large grid-connection requests do not represent equivalent amounts of executable data-center capacity. Texas, for example, reports more than 474 GW of ERCOT connection requests in August 2026. This figure represents requests for grid capacity and does not establish that 474 GW of data-center capacity is announced, financed, permitted, under construction, or scheduled for operation.

The National Data Center Pipeline Establishes the Scale of Planned Demand

The June 2026 Lawrence Berkeley National Laboratory (LBNL) United States Data Center Energy Usage Report: 2025 Update provides the national electricity-demand context for the U.S. data-center buildout. LBNL's Reference Case estimates 649 TWh of data-center electricity consumption in 2030, equivalent to 11.8% of total U.S. electricity consumption. The report also presents sensitivity scenarios in which data centers account for 9.5%–15.3% of U.S. electricity consumption in 2030. These figures represent an electricity-demand outlook rather than a project inventory, providing macro-level context for the scale of infrastructure required to support continued U.S. data-center expansion.

The August 2026 US Data Center Map tracker reports 1,420 U.S. data-center projects in its construction and planning pipeline, representing 318.6 GW of reported or announced power. This comprises 90.4 GW under construction across 378 projects and 228.1 GW planned across 1,042 projects. The planned segment therefore represents the larger share of the reported pipeline and identifies capacity that DCMap classifies as not yet under construction.

Under-construction and planned capacity represent two distinct stages of data-center development, and the distinction between them is central to this report. For executives and policymakers, the practical takeaway is that the U.S. data-center market contains a substantial development pipeline, while a smaller portion of this capacity has progressed into active construction.

Source: US Data Center Map, August 2026 national tracker.

The 2026 Delivery Test Faces a Significant Schedule-Risk Gap

Currence's February 2026 Data Center Outlook identified at least 16 GW of data-center capacity slated to come online in 2026 across roughly 140 projects, but only about 5 GW was under construction at the time of the outlook. Approximately 11 GW remained in the announced stage with no visible construction progress, despite typical data-center construction timelines of 12–18 months.

Currence also reported that 26% of expected 2025 capacity slipped, while another 10% of projects pushed back their commercial-operation dates. Based on that delivery record, Currence stated that it would not be surprising if 30%–50% of the capacity slated for 2026 were delayed. The Currence analysis therefore supports a schedule-risk range, not a cancellation range. Applying the 30%–50% delay-risk range to the 16 GW 2026 target produces an illustrative 4.8–8.0 GW of capacity potentially delayed.

U.S. Data Center Pipeline Remains Concentrated Across a Limited Number of States

The state data shows a highly concentrated pipeline, where Texas has the largest reported pipeline at 85.4 GW, followed by Virginia at 29.5 GW, Pennsylvania at 19.6 GW, and Utah at 18.7 GW. Together, these four states account for approximately 48% of DCMap's 318.6 GW national pipeline. Georgia, West Virginia, Illinois, New Mexico, Arizona, and Ohio form the next tier.

Concentration matters because project constraints are local. Transmission capacity, generation and fuel availability, interconnection processes, water availability, land-use approvals, local policies, and utility cost-allocation rules vary across states and counties. As a result, a national delay percentage can obscure substantial differences in the execution conditions facing individual projects.

For this reason, the report uses state-level pipeline power to measure the reported scale of the development pipeline, while relying on project-level evidence to assess actual deliverability risk. DCMap's pipeline-power figures can represent different underlying measures, including IT load, utility service, campus capacity, or another reported project measure. These figures should therefore not be interpreted as a standardized measure of IT load.

A large state-level pipeline does not necessarily indicate strong execution prospects. A state may have a large reported pipeline in which a substantial share of capacity has already progressed into construction, while another state may have a smaller pipeline in which a larger share remains at an earlier development stage. State-level pipeline size therefore provides the baseline for the analysis, while project-level evidence determines how much of that pipeline is actually deliverable. The full 52-state/territory dataset is available in the Project Dataset Explorer below.

Data Center Pipeline and Project Evidence, State by State

Every state combines the same underlying pressures — power, permitting, financing, and community response — into a different policy mix. The profiles below summarize the most evidence-rich states in the DCMap dataset.

Use the Project Dataset Explorer below to filter projects by state, power strategy, and status.

Project Spotlights

These three projects illustrate how delivery evidence actually accumulates in practice — and how it should be tracked across power, construction, economic activity, and compute delivery, rather than as a single "built / not built" flag.

Power Infrastructure: Grid, Behind-the-Meter, and Hybrid Models

Grid interconnection timelines remain a primary constraint on data-center delivery in many regions, prompting a growing share of developers to pursue behind-the-meter, onsite, or hybrid generation instead of waiting solely on utility interconnection queues. Texas provides a clear example of this shift: Microsoft/Pecos, Prometheus/Reeves County, and other Texas projects demonstrate the move toward behind-the-meter and hybrid power architectures as developers try to address grid-connection timelines while pursuing faster routes to large-scale power.

Onsite and hybrid systems improve the speed and flexibility of power delivery, but they also introduce their own infrastructure, permitting, and operating requirements — fuel supply and emissions permitting for gas generation, or fuel-cell/microgrid engineering for projects like Oracle's Project Jupiter. In practice, a project's power strategy is one of the clearest single indicators of how seriously to weight its stated timeline.

How Political and Regulatory Factors Affect Data Center Capacity

State-level regulatory responses are diverging, and each mechanism addresses the same underlying concerns from a different angle — infrastructure cost responsibility, financial risk allocation, and project approval:

• Texas is increasing scrutiny of data-center projects and their grid-connection requests.
• Arizona is strengthening large-load service agreements under a "Growth Pays for Growth" approach to new infrastructure costs.
• Pennsylvania is tying state support and permitting to responsible-development requirements, legally binding commitments, and local approval.
• Virginia is tightening large-load cost allocation and long-term service obligations.
• Georgia and other states are seeing more development decisions move into local zoning, water, and infrastructure processes.

Illustrative Capital Value of the Planned Pipeline

DCMap's 228.1 GW of planned (not-yet-under-construction) U.S. capacity represents a substantial potential capital commitment, though the actual figure depends heavily on the installed cost per megawatt assumed — which itself varies by region, power strategy, and project specification. The chart below applies three illustrative cost assumptions to size the range, rather than asserting a single point forecast.

How Is the Risk Classification Assigned?

The risk classification is an evidence-based assessment of project deliverability. It does not represent a probability of cancellation, a forecast of financial loss, or a judgment about the quality of a developer. The classification measures the strength of publicly documented evidence that a project is progressing toward physical delivery.

The assessment uses four execution gates, each evaluated using the most recent publicly available evidence for the project:

• Power — interconnection status, PPA/onsite supply, or grid agreement.
• Site and permitting — land control, air/water/land-use permits, zoning approval.
• Financing and customer commitment — secured capital, anchor tenant or hyperscaler commitment.
• Construction — documented site work, steel, equipment, or substation activity.

Projects with documented evidence across multiple gates receive stronger deliverability support; projects supported primarily by an announcement, with limited evidence across the other gates, receive weaker support. Evidence of a permit denial, project pause, customer withdrawal, financing failure, or other explicit execution constraint carries more weight than the mere absence of public information — the classification distinguishes between limited evidence and negative evidence. The strongest candidates for a delivery gap are therefore not simply the largest projects; they are projects with large announced capacity and limited evidence across multiple gates.

Source hierarchy used in this report

• Priority order: government/regulator sources rank highest, followed by company primary sources, then association/research organizations, then reputable news, then secondary trackers.
• National vs. state datasets: Currence/Sightline's 190 GW/777-project universe is used for the national announcement story; DCMap's August 2026 state table is used for the state-by-state reported pipeline. These are not mixed arithmetically, since they use different inclusion dates and coverage.
• Deliverable proxy: "under construction" is used as the strongest publicly visible near-term deliverability proxy; operational capacity is shown separately where available.
• Gap definition: state-level "pipeline power" is a development overhang, not a cancellation estimate. A project is not classified "at risk" unless evidence shows delay, pause, permit friction, financing stress, customer pullback, or political obstruction.
• 2026 schedule risk: Currence/Sightline's 30–50% delay estimate is applied only to the 16 GW of 2026-targeted capacity, not mechanically to every state.
• Dollar values: Sightline's 2025 median installed-cost benchmark of roughly $5.5M/MW is used as a sensitivity; all dollar outputs are labeled illustrative.
• Capacity types: IT load, critical IT, compute capacity, gross power, and leasable power are kept distinct — a generation project's MW is never added to a data center's MW.
• Ghost demand: electricity interconnection requests are not automatically data-center pipeline. Texas's >474 GW figure is shown as a grid-request signal, not counted as project capacity.

Project Dataset Explorer

Filter the underlying state-level pipeline dataset (52 states/territories) and the project-level evidence dataset (18 flagship projects) compiled for this report.

0 records shown

Sources: US Data Center Map, State Dataset (Aug 2026 update) for the state table; NMSC-compiled project evidence (company disclosures, state/local filings, regulatory dockets) for the project table — see Methodology for the source hierarchy applied.

Conclusion: Assessing Data Center Deliverability Beyond Announced Capacity

The state-level and project-level evidence shows that deliverability is increasingly shaped by execution conditions at the project and local level. Texas is tightening verification of large-load interconnection requests, Virginia is increasing customer responsibility for infrastructure costs, Arizona is strengthening protections for existing electricity customers, Pennsylvania is applying additional requirements to supported projects, and local governments in states such as Georgia are imposing zoning, water, and development controls. At the project level, developments such as Microsoft's Pecos campus, Meta's Richland Parish campus, and Oracle's Project Jupiter demonstrate that large-scale projects advance through multiple stages of power development, construction, infrastructure investment, and operational delivery.

The central conclusion is that the U.S. data-center boom should not be measured solely by announced gigawatts. The more meaningful measure is the amount of capacity supported by documented progress across power, site and permitting, financing and customer commitment, construction, and ultimately energization. The accompanying project-level workbook applies this framework to distinguish announced pipeline capacity from capacity with stronger evidence of delivery, and provides a structured basis for tracking changes as new project evidence emerges.