Why AI Data Center Projects Face Years of Delays After Approval

extract

Why AI Data Center Projects Face Years of Delays After Approval New PJM data reveals AI infrastructure projects now spend more time waiting after interconnection approval than in the queue itself. AI infrastructure projects entering service in 2025 took an average of more than seven years to reach operational status, according to PJM Interconnection data – but the biggest delays are no longer occurring within the interconnection queue. The bottleneck has shifted downstream. Transmission buildouts, substation capacity, and strained supply chains are now the primary obstacles to energizing projects. According to PJM, projects spent an average of more than three years reaching an interconnection service agreement, and another four years waiting to come online after approval. Turning approved interconnection rights into energized megawatts hinges on overcoming these challenges, reshaping timelines for the next generation of data centers. As demand for hyperscale AI campuses accelerates, these downstream issues highlight the growing complexity of energizing large-scale infrastructure. “That confirms what we have been trying to emphasize – the issues outside of the queue are the biggest obstacle we face to bringing projects online,” Jeff Shields, PJM senior manager of external communications, told Data Center Knowledge in an email. The figures complicate a growing narrative that interconnection queues alone are driving delays in large AI infrastructure projects. Recent analysis has cited timelines of approximately eight years for projects reaching operation in PJM territory. Shields said those figures often mischaracterize the transmission company’s role by including construction and post-approval development work in the queue timeline. “A project is finished with the PJM process and requires nothing further from PJM once it signs an Interconnection Service Agreement,” Shields said. Queue Reforms Aim to Shorten Timelines PJM has overhauled its interconnection process in recent years, replacing its first-come, first-served structure with a cluster-based review system intended to reduce backlog and accelerate approvals. Under the reformed process, PJM said new Cycle 1 projects are expected to receive interconnection agreements within one to two years, depending on system impact. The organization said it has processed more than 170,000 MW of generation requests since 2023 and currently has about 30 GW of projects left to process from transition cycles. PJM recently received applications representing roughly 220 GW of proposed projects for its next interconnection cycle, though the organization said that figure remains preliminary pending application review. PJM’s 2026 load forecast projects that summer peak demand will rise from roughly 154 GW in 2025 to nearly 210 GW by 2036, driven primarily by data center expansion and electrification. The scale of new large-load requests is already accelerating. PJM said it received 95 large-load adjustment requests totaling around 54 GW through November 2025. That demand is arriving in massive blocks of load, with utilities and grid operators evaluating AI campuses requiring hundreds of megawatts of capacity and, in some cases, gigawatt-scale deployments. The Bottleneck Moves Downstream PJM data presented in January 2026 showed permitting accounted for 29% of milestone change requests for projects under construction or development, followed by “other” delays at 28% and supply chain delays at 23%. PJM said “other” delays included issues tied to EPC procurement, equipment availability, construction delays, and land ownership. Large power transformers used in substations and transmission infrastructure are facing sharply longer procurement timelines as utilities and developers compete for limited manufacturing capacity. Research from Wood Mackenzie found average transformer lead times increased from roughly 50 weeks in 2021 to about 120 weeks in 2024, with some large substation and generator step-up transformers ranging from 80 to 210 weeks. Wood Mackenzie projects the US data center electrical equipment market will expand from roughly $20 billion today to $65 billion by 2030, as hyperscale AI construction accelerates. Ben Boucher, principal analyst at Wood Mackenzie, said developers are increasingly competing with utilities for the same pool of electrical equipment supply as AI-related demand accelerates. Boucher said substation transformer lead times averaged roughly 140 weeks in 2023, increased to about 150 weeks in 2025, and now exceed 160 weeks in 2026 based on the firm’s latest data. “Time to market is one of the most important aspects for developers, and if equipment lead times continue to rise, project delays will become more frequent,” Boucher said. Projects are increasingly securing interconnection agreements before the infrastructure needed to energize them can be delivered. As of January 2026, PJM reported more than 21 GW of projects in engineering procurement status and another 8.2 GW under construction. Transmission Delays Reach Into the Next Decade Real-world transmission delays are already affecting AI-linked infrastructure projects. In a March 31 filing with the Federal Energy Regulatory Commission (Docket No. ER26-2028), Constellation Energy said the restart of the former Three Mile Island Unit 1 nuclear plant would require significant transmission-related work before the facility could become fully deliverable within PJM. The company warned the facility may not be “fully deliverable” until those upgrades are completed. Constellation has said the restarted plant will supply power under a long-term agreement with Microsoft. The filing shows how transmission schedules and equipment availability are increasingly dictating when AI infrastructure can actually come online. For developers racing to deploy AI infrastructure, the challenge is no longer just securing land, financing, or interconnection rights, but also the speed with which utilities, contractors, and equipment suppliers can deliver power to the site. Increasingly, that timeline begins after the queue ends.