A regulatory fight in Georgia and a turbine shortage traced to Ohio show how AI infrastructure spending is reaching companies that have nothing to do with data centers.
In December 2025, the Georgia Public Service Commission approved a plan letting Georgia Power add 9,985 megawatts of generation by 2030. Three years earlier, the utility’s own forecast called for 400 megawatts. The commission says roughly 80 percent of the new demand comes from data centers.
Five months later, the same commission opened a forensic investigation into whether its cost-allocation rule for those customers is actually working. The rule, adopted in January 2025, requires large-load customers to pay for the transmission and distribution capacity built to serve them. Whether Georgia Power is applying it as designed is now an open question the commission is examining rather than a settled fact.
None of this involves oil or gas. But oilfield services companies with midstream customers in Georgia, Texas, and Virginia are discovering that state regulators now have less time for pipeline approvals, gathering permits, and rate cases, because AI infrastructure has taken over the docket. Master service agreements written before this backlog rarely account for regulatory delay as a distinct category, separate from weather or force majeure, which leaves the question of who eats the cost of a stalled permit sitting in a gap most contracts never anticipated.
The Queue Problem
Grid operators call it an interconnection queue, the list of projects waiting for permission to plug into the transmission system. Lawrence Berkeley National Laboratory’s 2026 data puts the median wait, from application to commercial operation, at more than five years for projects that get built at all. Most don’t. In ERCOT, 60 percent of projects that entered the queue by 2020 never reached an interconnection agreement. In PJM, the figure is 76 percent.
PJM’s backlog got bad enough that the grid operator stopped accepting new applications for several years. It reopened in 2026 with a redesigned process meant to clear more than 300 gigawatts of stalled requests. ERCOT’s large-load queue, meanwhile, went from 63 gigawatts at the end of 2024 to 226 gigawatts a year later. Data centers account for 77 percent of that, most targeting a 2030 connection date that the failure rates above suggest most won’t make.
For a service company pricing a job off a customer’s stated in-service date, that history matters more than the date itself. An operator’s 18-month timeline for a new facility carries roughly even odds of slipping by a year or more once queue delays are factored in, and the odds worsen with project size. Tie milestone payments and mobilization schedules to a documented interconnection agreement or permit approval, not a projected completion date the customer doesn’t control.
The Equipment Fight
Gas turbine prices have risen 195 percent since 2019 and are on track to hit $600 per kilowatt by the end of 2027, according to Wood Mackenzie. Global orders now total 110 gigawatts against 60 to 70 gigawatts of manufacturing capacity. GE Vernova started 2025 with 46 gigawatts on contract and ended the year at 83. Siemens Energy is running plants around the clock and carrying a backlog of 136 billion euros, the largest in the company’s history. Mitsubishi Power is sold out through 2028.
Wood Mackenzie’s Aurora Tenorio attributes the bottleneck to more than turbine assembly: specialized labor shortages and component delays in hot-section manufacturing are compounding the problem. Transformers face separate lead-time constraints. The engineers and procurement staff being pulled toward gas-fired power projects for data centers are drawn from the same labor pool that oilfield services companies use for completion crews and pipeline construction.
The scale involved varies. The Portsmouth Powered Land Project, a $33 billion, 9.2-gigawatt facility proposed for southern Ohio in February 2026, could require 24 to 30 heavy-duty turbines on its own, an unusually large order that illustrates the upper end of demand rather than the norm. Most data center campuses need a fraction of that. But even smaller projects are now competing for the same manufacturing slots and skilled labor that oil and gas projects used to source without this kind of competition. Any fixed-price bid signed this year should include a price escalation clause indexed to a named equipment or labor benchmark, not an open-ended cost adjustment provision that invites dispute later.
Georgia’s forensic review is due to conclude by the end of 2026. PJM’s redesigned queue has yet to prove it can clear its backlog faster than the system it replaced. Turbine manufacturers are adding capacity, but not on a timeline that closes the gap between 110 gigawatts of orders and 70 gigawatts of supply anytime soon.
