Currence Forecasts Up to 50% Delay in 2026 AI Data Center Builds
Mounting power grid constraints and construction bottlenecks threaten to slow the rapid expansion of gigawatt-scale compute infrastructure.
A new market outlook report from energy intelligence firm Currence warns that between 30% and 50% of the large-scale data center capacity scheduled to come online in 2026 will likely face delays. The study, which tracks 777 announced AI factories and hyperscale facilities exceeding 50 megawatts, highlights a growing divergence between planned infrastructure and actual construction progress. Mounting electrical grid limitations, local regulatory opposition, and labor shortages are increasingly stalling projects, forcing operators to pivot toward independent, behind-the-meter power solutions.
Key details
According to the Currence report, developers have announced approximately 190 gigawatts (GW) of future data center capacity since 2024. Of this total, roughly 16 GW of capacity was scheduled to begin operations during 2026. However, the study found that only about 5 GW is currently under construction, leaving 11 GW—nearly 70% of the projected 2026 pipeline—stuck in the planning stages with no visible construction progress despite typical build cycles of 12 to 18 months.
The report also reveals that project timelines are becoming highly unreliable across the industry. In 2025, more than 25% of expected capacity missed its projected completion date, while another 10% of projects quietly pushed back their commercial operation schedules. This bottleneck is heavily driven by power availability; while projects utilizing on-site generation or hybrid systems represent fewer than 10% of announced facilities, they account for nearly half of the total announced capacity.
In addition to physical infrastructure delays, massive financial pressures are mounting. Hyperscalers have accrued substantial debt to finance this buildout, with Bloomberg estimating outstanding AI data center debt has crossed $500 billion, while Nikkei Asia reports the five largest tech giants have accumulated up to $1.65 trillion in total debt over the past five years, including hundreds of billions in off-balance sheet liabilities.
Why this matters
The Currence report underscores a critical physical constraint on the AI boom: artificial intelligence cannot scale without massive physical infrastructure, and the physical world cannot keep pace with software projections. This delay directly impacts the timeline for training next-generation foundation models and deploying enterprise-scale AI tools. It also demonstrates that "speed to power" has surpassed chip availability as the primary competitive bottleneck in the AI race, driving up capital expenditures and creating significant financial risk for developers who have over-committed on unbuilt capacity.
Context
Over the past two years, the rapid growth of generative AI has triggered an unprecedented surge in data center power and water demands, leading to severe strain on public utility systems. Across the United States, utility companies have struggled to integrate these massive loads, resulting in soaring electricity bills for residential ratepayers and rising community pushback. This friction has prompted a flurry of state-level regulations and city-level moratoriums on new data centers. In response, hyperscalers like Google and Amazon are increasingly bypassing public grids by acquiring renewable energy pipelines, investing in direct solar generation, and deploying utility-scale battery energy storage systems.
What happens next
To bypass utility interconnection delays and mitigate public backlash, developers are expected to rapidly expand their investments in off-grid, behind-the-meter generation. The industry will likely see a surge in the deployment of mobile battery storage systems, on-site natural gas turbines, and fuel cells to power facilities independently. Additionally, as the distinction between announced and executable projects becomes more stark, utilities and equipment suppliers will need to implement more rigorous screening processes to separate speculative proposals from projects with secured power and land contracts.
Source: Network World Published on AI Usage Global, author: AUG Bot
