As the data center industry rushes to meet demand for AI compute, new projects have been announced at an ever-increasing rate. The research firm Wood Mackenzie, however, reported a slowdown toward the end of 2025. Although 25 GW of new project capacity was added to the US pipeline in the last quarter of the year (an impressive figure by 2023 or 2024 standards) this figure was only around half of the exceptionally high 49 GW added in the previous quarter.

Looking at capacity under construction, commercial real estate firm CBRE reported that projects underway in North America's primary markets fell to 6 GW at the end of 2025, down from 6.4 GW at the end of 2024.

This reduced pipeline will slow capital expenditure growth among the largest developers, including hyperscalers, for the first time since 2023, reducing it by 58% year-on-year to an estimated US$94 billion in 2026.

Why are fewer projects being announced and built? Colocation capacity is becoming scarce (vacancy rates are a record low of 14%, reports CBRE) and funding is plentiful (around $160 billion was invested in US data centers in 2025, which is 30% more than in 2024, according to the US Department of Commerce). Three major contributing factors were outlined at the Uptime Network Americas Spring Conference 2026 in Dallas in March.

Connection delays

Data center power consumption, once relatively small compared with other sectors, has expanded to become one of the most significant factors in power demand growth in several states (see Table 1). In Texas, data center power demand is expected to grow by 230 GW between 2027 and 2035, reaching around 258% of current peak electricity demand. In Utah, Wisconsin and Illinois, data centers are projected to add 140%, 35%, and 62%of current peak demand, respectively. Globally, data centers are expected to account for 20% of power demand growth through 2030, according to the International Energy Agency (IEA).

Table 1 Data center demand growth compared with peak demand in some US states

This level of power cannot be provided quickly or easily, with utilities — and their equipment suppliers — having to rapidly scale up their operations to meet demand. Grid connection delays are now the most significant bottleneck in development timelines, with large customers facing waits of four to 10 years, and sometimes longer.

To avoid these delays, data center developers are turning to on-site power generation. However, in practice, behind-the meter power generation is constrained by complex local regulations and resistance from local communities.

Legislation and opposition

During the first 6 weeks of 2026, at least eight US projects were canceled or put at risk of cancellation (see Data center cancellations on the rise as public opposition grows). A rising tide of public opposition, combined with the prospect of tighter state regulations has created significant difficulties in securing project approvals.

Local opposition is based on perceptions of high environmental impact of data centers and increased consumer power costs, as well as the lack of sufficient local economic benefits to justify the often-generous tax incentives offered to operators. Critics also object to non-disclosure agreements (NDAs) with local governments and utilities, which often prevent scrutiny of large data center projects in their early stages.

Data center opposition is increasingly coordinated and informed by national bodies. In early 2026, 230 advocacy groups urged the US Congress to impose a federal moratorium on data center construction pending tighter regulations. A national pause is unlikely, but moratoria have been approved in several cities and counties and are being considered at the state level.

On-site power generation will require legislative changes (it is not permitted in some states), and will always remain heavily regulated (see Regulations for behind-the-meter power are emerging).

Some states and public utility commissions are now regulating the impact of data centers on the grid, requiring actions that include:

• Demand flexibility. Data center operators are expected to reduce demand by switching to other power sources or scaling back operations on demand.
• Ride through. Facilities are expected to maintain grid connections during faults, to protect the grid from simultaneous actions by their UPS systems. (see Power companies act to stop data center-induced blackouts).

Supply chain issues

Grid connection delays are merely another aspect of near-universal supply chain stress. Large data centers are being built in phases, but supporting industries cannot scale capacity in the same way.

Equipment that was previously relatively easy to source, such as engine generators, now have lead times exceeding a year. For equipment required for on-site power generation, such as gas turbines, lead times can extend past 2 years — underlining the reality that behind-the-meter power is not a simple solution to connection delays.

Scaling up the manufacturing and supply of equipment takes time, and manufacturers will be reluctant to invest in greater capacity unless they are confident that demand will remain high over the next 10 years. Perceptions of a possible future reduction in demand only compounds supply chain uncertainty. In addition, simply training up enough staff to manage this surge in new capacity is an acute issue (see Survey highlights industry staffing crisis).

The pipeline figures for US data center development confirm what has long been evident: developers in the US (and elsewhere) are struggling to deliver the capacity they have promised.