NEW YORK — What happened
Regional transmission organizations and utility companies report that interconnection queues for new multi-gigawatt AI data centers are severely backlogged. Waiting times to connect large-scale computing facilities to high-voltage transmission lines now routinely exceed five years in primary tech hubs.
Why it matters
Physical electricity availability has officially surpassed silicon supply as the primary constraint on expanding large-scale AI training clusters. Without immediate power access, billions of dollars in procured accelerators risk sitting unpowered in warehouses.
Deep dive
Templated data center designs that once required standard 100-megawatt connections are now scaling toward gigawatt campuses to support dense GPU arrays. Local electrical grids, designed for decentralized suburban loads, cannot handle sudden localized spikes without destabilizing regional voltage. Consequently, AI infrastructure developers are bypassing public utility queues by investing directly in behind-the-meter power solutions, including solar-storage microgrids and advanced nuclear small modular reactors (SMRs).
Report check (claims vs what is verified vs still rumor)
Corporate announcements claim that private nuclear power purchase agreements will supply fully operational carbon-free data centers by 2027. Energy regulators verify that while land leases and corporate agreements are signed, actual regulatory approval and construction timelines for SMRs point toward the 2030s. Rumors of immediate grid bypasses via experimental fusion power remain unsubstantiated.
Open questions
Who absorbs the infrastructure upgrade costs when large tech campuses require massive high-voltage substation expansions?