Only half of planned data center capacity for 2026 is under construction

What happened?

A report from investment bank Jefferies, shared with The Register, reveals that the data center capacity actually under construction in the United States is far lower than announced. Of the 24 GW promised for 2026, only half (12 GW) has started construction. For the 2027-2028 period, up to 80% of planned capacity has not yet begun construction. Causes include zoning issues, permits, electrical interconnection, energy access, labor shortages, and delays in signing commercial contracts. Jefferies also highlights that double-counting of applications by hyperscalers to multiple utilities artificially inflates total announced capacity figures.

Why is it important?

This mismatch between announcements and reality has direct implications for the AI and cloud industries. The demand for computing to train and run AI models has skyrocketed data center growth forecasts. However, if infrastructure does not materialize at the expected pace, bottlenecks could occur, making AI services more expensive or delayed. The report notes that energy availability is the main bottleneck, with delays of up to seven years for grid connections. In fact, the US Secretary of Energy has urged the Federal Energy Regulatory Commission (FERC) to implement new rules to streamline the process for customers like data centers. Additionally, the labor market for data center construction is limited, slowing the pace of work. Jefferies estimates that only 50% of announced capacity for 2026 will be completed on time, and for 2027-2028 the figure could be even lower.

Market consequences

Jefferies estimates that the realistic pace of new capacity is 15-20 GW per year, compared to the 40+ GW some forecast for 2027-2028. This suggests that investor expectations may be overvalued and that the data center market could face a correction. Operators are migrating to regions with better interconnection and permitting conditions, such as Texas, which added 14 GW of new announced capacity in the second quarter of this year alone. Hybrid and behind-the-meter models are also being adopted to circumvent grid limitations. In the hybrid model, data centers first take all available energy from the grid and then resort to on-site generation. This reflects a trend toward self-generation and location in areas with fewer regulatory hurdles. The Jefferies report also notes that the project pipeline is shifting toward regions like Texas, Ohio, and Virginia, where conditions are more favorable.

What readers should know

• Not all announced capacity will be built: investors should rely on purchase agreements, permit progress, and realistic timelines.
• Energy availability is the main bottleneck, with delays of up to seven years for grid connections.
• The labor market for data center construction is limited, slowing the pace of work.
• Mitigation strategies include on-site generation and location in areas with fewer regulatory hurdles.
• Double-counting of utility applications artificially inflates planned capacity figures.
• Only 15-20 GW of new capacity per year is realistic, compared to the over 40 GW some forecast.

"Announced capacity should not be considered a reliable way of evaluating data campus load growth" — Jefferies

In summary, the Jefferies report highlights a significant gap between data center capacity announcements and actual construction, which could have consequences for the AI and cloud supply chain as well as investor expectations. The combination of regulatory, electrical infrastructure, and labor issues suggests growth will be slower than anticipated, forcing the industry to adapt with alternative models and strategic locations.