Why FAST-41 Now Covers AI Data Centers and Copper

extract

Why FAST-41 Now Covers AI Data Centers and Copper Federal permitting now covers hyperscale campuses, critical minerals, and transmission in a single framework. The federal government is increasingly treating AI growth as an industrial infrastructure problem rather than simply a technology boom, folding hyperscale data centers, transmission, and critical materials into the same permitting and planning process. FAST-41 – the federal framework created to streamline and coordinate environmental reviews for major infrastructure projects – now explicitly covers AI-linked developments. The Federal Permitting Improvement Steering Council (FPISC), which oversees FAST-41, has granted FAST-41 coverage to Alaska’s Arctic copper and critical minerals project, extending the framework into supply chains that underpin substations, transformers, switchgear, and transmission systems used by hyperscale campuses. Coverage expansion follows earlier FAST-41 changes that added "Artificial intelligence and machine learning" and "High-performance computing, advanced computer hardware and software" to eligible categories and brought a QTS campus in Richmond, Va., into the framework as the program’s first covered data center project. The Richmond campus spans seven buildings and roughly 3.2 million square feet. Shahid Mahdi, director at regulatory intelligence firm EnerKnol, said hyperscale campuses are beginning to reshape long-term utility planning assumptions. In some cases, he said, a single AI campus can demand as much electricity as roughly one million homes. For hyperscale developers, the implications extend well beyond permitting policy: delays in transmission construction, transformer manufacturing, and grid equipment deliveries are increasingly affecting campus construction schedules, power delivery timelines, and expansion plans across major data center markets. FAST-41 in Practice for AI Development Federal filings tied to the QTS project show how directly hyperscale development is now entering the FAST-41 process. QTS told federal officials the project would support AI-related workloads and noted the company was simultaneously constructing hyperscale deployments at nearly 4 GW. Taken together, the QTS and Arctic projects show how AI expansion is pulling data centers, transmission systems, electrical equipment manufacturing, and raw material supply deeper into the same infrastructure planning process. Jigar Shah, former director of the Department of Energy’s Loan Programs Office, told Data Center Knowledge that the expansion reflects a broader change in how policymakers now view AI development. For much of the last decade, he said, AI was treated as a software-and-services story that technology companies could manage alongside electric utilities and state regulators. “The FAST-41 expansions … reflect a different understanding: that AI infrastructure is a physical industrial system with the same supply chain vulnerabilities and strategic dependencies as any other major infrastructure buildout,” Shah said. Why Copper and Equipment Bottlenecks Matter The Arctic project centers on copper production, a material used extensively in substations, transformers, switchgear, transmission, and backup power equipment supporting hyperscale campuses. “The United States currently imports a substantial share of its copper supply from foreign nations – a strategic vulnerability that FAST-41 could help alleviate,” Trilogy Metals CEO Tony Giardini said following the Arctic project announcement. Constraints are already visible in the grid and equipment markets. NERC, the North American Electric Reliability Corporation, has projected that summer peak demand could rise by 224 GW over the next decade, driven largely by data center growth. Shah said electrical infrastructure shortages are already slowing projects. “Transformers are running 18- to 24-month lead times,” Shah said. “Wires and cables have inflated 152% since 2019. Switchgear is up 77%. Nearly half of US data centers planned for 2026 are facing delays or cancellations.” Industry analysis from Wood Mackenzie estimates that US demand for generator step-up transformers has climbed 274% since 2019, while power transformer demand has risen 116%, pushing the market into supply deficits and higher prices. Sheets, or cathodes, of 99.99% pure copper sit in stacks outside a solvent-extraction plant awaiting shipment. (Image: Getty) Regulatory Responses and Risk Allocation Regulators are starting to adapt by imposing new financial protections around large-load development. In Michigan, state regulators recently required DTE Electric to use a 19-year take-or-pay structure with large-load customers to reduce the risk of stranded infrastructure costs. As campus sizes climb into the gigawatt range, utilities and grid operators are increasingly treating hyperscale developments less like ordinary commercial loads and more like major pieces of electrical infrastructure. Schedule risk is often acute. In several major markets, developers can complete a large data center shell long before utilities can deliver new transmission capacity or secure high-voltage transformers. Northern Virginia has become one of the clearest examples, with transmission construction and interconnection timelines stretching years beyond planned deployment schedules. ERCOT, the grid operator in Texas, has also begun formalizing work around large AI-related loads. A newly approved Large Load Working Group charter identifies low-voltage ride-through criteria for data centers, rapid large-load ramping behavior, harmonics and oscillations, and co-located on-site generation as emerging operational concerns for the Texas grid. Architecture: Gigawatt Campuses vs. Distributed Growth The scale of compute is intensifying the debate over where and how to build. “AI compute stock is doubling every seven months,” Shah said. “A new grid interconnection in a constrained market like Northern Virginia or PJM takes seven to eight years on average.” Shah argued the larger problem is architectural. “The US is currently building AI infrastructure on gigawatt-scale campuses even though inference can be done at units that are 100 MW and below, which are much easier to bring onto the grid,” he noted. Mahdi said the mismatch between AI deployment cycles and grid infrastructure timelines is pushing developers toward private power strategies. “AI companies operate on product cycles measured in months, while energy infrastructure operates in years or decades,” he said. Developers have pursued behind-the-meter and co-located generation as interconnection timelines stretch further into the next decade. Mahdi pointed to projects, including the GW Ranch development in Texas and Meta’s Socrates campus in Ohio, as examples of attempts to bypass congested interconnection queues with dedicated on-site generation. Shah cautioned that these approaches may prove difficult to sustain at hyperscale over the long term. “The behind-the-meter track is a dead end, and folks are starting to realize it,” Shah said. “The grid-connected track solves a long-term cost and reliability problem.” With equipment backlogs, grid queues, and regulatory risk reshaping build schedules, utilities, grid operators, and hyperscale developers are racing to reconcile AI deployment timelines with an electrical system that expands far more slowly.