The Turbine Shortage Is the New Interconnection Queue: Procurement in the AI Era
Written by Bruno P. Allaire
The buildout's second bottleneck isn't the grid, it's the factory. Turbine and engine lead times now stretch past the interconnection queue itself, and the scarce asset of this era is a manufacturing slot. Projects that solved their power problem will still die waiting for iron.
Share this perspective
The Second Queue
The industry has spent two years learning about the interconnection queue. Load can't get served because the studies take half a decade, and the story has been told often enough that boards now open siting conversations with it. But something quieter has been happening to the projects that did solve their power problem, the ones that bridged behind the meter or co-located generation and cleared the grid entirely. They are arriving at the factory and discovering a second queue, and the second queue is longer than the first.
Turbine and reciprocating-engine lead times, which ran six to twelve months in a normal market, have stretched to three years and beyond for the larger frames. Order books at the major OEMs are oversubscribed well into the back half of the decade: GE Vernova closed its most recent quarter with 116 gigawatts of gas equipment backlog and paid slot reservations, and is taking reservations for delivery as far out as 20311, while Siemens Energy reports lead times of three years or more across its gas turbine business2. The dealer market for the mid-size units has gone from a phone call to a waiting list, and the waiting list assumes you already have a deposit down. A sponsor who cleared interconnection this year and placed an equipment order the same month may energize a campus in 2027 and take delivery of the turbine in 2028. The project didn't fail. It just stopped.
This inversion deserves more attention than it gets. For most of the last decade, a data center's critical path ran through permits and fiber and, lately, the utility. Now, for a growing share of serious projects, the critical path runs through a foundry in the Midwest or a test stand in Europe. The scarce asset of the AI buildout is no longer megawatts. It's a manufacturing slot.
Why Manufacturing Doesn't Scale Like Software
The instinct in technology circles is that capacity follows demand: announce the load, and the supply chain will build toward it. Heavy manufacturing doesn't work that way, and the reasons are worth stating because they determine how long the shortage lasts. A gas turbine is not a chip design. It is an assembly of large castings and forgings that only a handful of foundries on earth can pour, moved through machining cells and heat-treatment ovens that were sized for a market that peaked a decade ago.
The industry let that capacity atrophy on purpose. After the last buildout wave, the OEMs consolidated their product lines, retired the older frames, and shrank their supplier networks to match a gas-turbine market that everyone assumed was in secular decline. The specialized welders, the blade-coating shops, the test stands capable of running a full frame, all of it thinned out. Nobody is proud of it now, but it was rational then, and the fact that it was rational then is exactly why it can't be undone quickly now. Rebuilding a foundry line takes years and a demand signal durable enough to justify the capital. A model release cycle is not that signal, and the foundries know it.
Add the labor problem, which is the quietest constraint of all. The skilled trades that build this equipment skew toward retirement, and the training pipeline for turbine-grade machinists and welders was allowed to thin along with everything else. Demand did not create the workers. So while the headlines quote gigawatts of announced load, the factory floor is working through the same arithmetic it always has: how many units can physically leave the dock this year, regardless of how many purchase orders are pinned to the wall. For scale, GE Vernova, the largest western manufacturer, produces on the order of twenty gigawatts of gas turbines annually for every customer on earth, while a single year of announced AI-era load now arrives in multiples of that3.
What a Slot Is Worth
Here is the consequence, and it changes how these projects should be underwritten. A position in an OEM order book is no longer a line item in a procurement file. It is a financial asset. The deposit, the delivery date, the cancellation and assignment rights, and the escalation protections in that purchase agreement now carry value the way a queue position or an offtake contract does, because they are the difference between a project that exists and a project that is a rendering.
The market is starting to price it that way, quietly. Slot assignments and order-book positions are being shopped, with paper moving between sponsors who placed orders early and those who arrived late with money and no iron. Smart buyers are beginning to diligence the purchase agreement with the same scrutiny they give the reserve report or the offtake: is the delivery date fixed or indicative, what penalties actually attach to it, can the position be transferred, and what happens to it in a default. I have underwritten enough energy paper to know what that shift means. When a document starts getting read like a security, it has become one.
The sponsors who placed orders in 2023 and 2024, before the lead times made the front pages, hold something that appreciates while everyone else studies. That is not a metaphor. A firm slot at a reasonable price, with delivery inside the demand window, is worth more today than the same slot was worth at signing, and the gap is widening monthly. Early procurement has quietly become the highest-return move a power sponsor made this decade, and most of them made it for boring, operational reasons before anyone called it strategy.
Vendor Financing, the Late-Cycle Signal
Then there is the signal I flagged in my funding piece as deserving its own treatment: the OEMs are starting to finance their own sales. When a manufacturer extends payment terms, offers leases, or wraps vendor financing around a turbine sale, it is doing two things at once. It is removing the last excuse a marginal buyer has, and it is protecting its own order book by making the order book bigger. Both tell you something about where we are in the cycle.
The historical echo is worth knowing, because this pattern has played out before in American heavy industry. The locomotive builders financed the railroads' purchases through the competitive mania of the mid-century, and the telecom equipment vendors financed the carriers through the late 1990s. In both cases the equipment maker eventually held receivables against customers whose business case depended on the same demand curve the equipment was serving. Vendor financing is what a seller does when demand is strong but the buyer's balance sheet is the weak link. It is a symptom of a peak, not of a trough, and it deserves a cold look precisely because it makes the easy deals easier.
None of this says the demand isn't real. It says the financing structure is migrating from the disciplined end of the market toward the speculative end, which is what happens in every buildout. The projects taking vendor financing today to stretch a thin capital stack are the ones to watch at the first soft patch, because the equipment will have a mortgage on it, and the mortgage will be held by the same party that has the delivery date. A sponsor in that position has given up the two things this era says you should never give up at the same time.
Underwriting Equipment Like an Energy Asset
So what does disciplined procurement look like when the equipment is the scarce input? The same standards energy finance has always applied to every other input, applied to iron. A delivery date backed by something stronger than a salesperson's optimism: liquidated damages that actually compensate the delay, milestone payments tied to physical progress rather than calendar quarters, inspection and witness-test rights at the factory, and a purchase agreement that names, in advance, what happens if the date slips by six months or a year.
The mid-market sponsor has one structural advantage here, and it isn't size. It's the refurbished and second-hand market. While the majors fight over new frames, there is a real population of late-life turbines and low-hour reciprocating units coming out of decommissioned plants, power barges, and deferred utility projects, and the sponsors who know how to buy an asset with an operating history are often delivering power years before their competitors' new iron clears the dock. That is where my own reflexes were formed, too. In the trough years, running a small-cap oil and gas company, we had more leverage than money, and I spent weeks negotiating equipment deliveries and refurb terms with vendors who had empty order books and full warehouses. You learn, permanently, that the purchase order is a deal to be underwritten, not a form to be signed. The market has now inverted, the vendors hold the leverage, and the discipline is the same one, just pointed the other way.
The honest caution is that the second-hand route carries its own diligence load: residual life assessments, rotor and combustor inspections, emissions retrofits to match current permits. Done carelessly it is someone else's problem being purchased at a discount. Done properly it is the fastest route to energized iron this market offers, and the operators who built that sourcing muscle before the shortage made it fashionable are, once again, the ones compounding quietly.
The Modularity Answer
Which brings me to the design decision behind our own venture, because procurement strategy and product architecture are the same question at this scale. It is why we formed Defiant Power Solutions around standardized, modular units rather than chasing the scarce big-iron slots. A single 450-kilowatt reciprocating block, parallelable to utility scale, is not a compromise on capacity. It is a hedge on the supply chain. The small standardized units are built in volume, shipped by truck, and staged in inventory, which means the lead-time exposure that kills a frame-order project is measured in weeks for us, not years.
The economics of modularity read like procurement insurance. Scale arrives in increments, so a campus can match its generation to the actual load ramp instead of pre-buying a decade of capacity on a promise. A delayed unit delays a unit, not a project, because the architecture doesn't collapse when one delivery slips. And the fleet distribution means the maintenance program, the parts inventory, and the operator training all standardize across every site, which is exactly the kind of boring compounding that separates operators from announcers.
I don't pretend the answer is modularity for everyone. The hyperscale campuses with firm fuel and investment-grade balance sheets will get their frames, and some of them will get them early. But for the mid-market, and for every sponsor whose project dies if the iron is late, the architecture decision is the procurement decision. Designing around the constraint beats negotiating against it.
The Window
So how do you read this market from the outside? The same way you read every other cycle: by the paperwork, not the press release. The utilities are already running that arithmetic themselves: Exelon cut its high-probability data center pipeline from roughly eighteen gigawatts to eleven in a single quarter, and Texas paused to audit an interconnection queue that had swollen to about 474 gigawatts of requests, some ninety percent of it data centers4. The interconnection filings and the fuel contracts tell you whether a project can be built. The order book tells you whether it will be. A sponsor holding firm delivery positions with penalty-backed dates is a different credit than a sponsor with a letter of intent and a rendering, no matter how similar their announcements read.
The window itself has a shape, and the shape is familiar from the last article's funding gap. The equipment shortage will eventually ease, the way every manufacturing bottleneck eases when capacity, prices, and demand find their new equilibrium. But that easing is years out, and the capital deployed in the meantime locks in the winners of the era. The slot-holders, the refurbishment operators, and the sponsors who designed their architecture around the constraint will deliver campuses through the shortage. Everyone else will deliver explanations.
The correction, when it comes, will sort them quickly, the way corrections always do. The speculators who ordered iron against announced load that never materialized will be assigning slots at a discount, and the operators will be there to buy them. That, too, is a pattern the trough taught me: the shortage punishes the unprepared first, and then, at the turn, it hands the prepared their best entry prices. The turbine queue is the interconnection queue's harder brother. Treat a manufacturing slot the way you'd treat a queue position or a lease: as the asset it has become.
Notes & Sources
- 1.GE Vernova closed its second quarter of 2026 with 116 GW of gas power equipment backlog and slot reservation agreements, up from 83 GW at the end of 2025, and is taking reservations for delivery as far out as 2031. Utility Dive, "GE Vernova gas turbine backlog climbs to 116 GW," July 2026
- 2.Siemens Energy reported gas-turbine lead times of three years or more alongside a 69 GW gas services backlog in its fiscal third-quarter 2026 results, and industry reporting puts average lead times for a new combined-cycle plant at roughly five years, up from about three and a half in 2023. Utility Dive, "Siemens Energy's gas turbine backlog nears 70 GW as company expands manufacturing," August 2026; OilPrice, "US Power Boom Triggers Global Gas Turbine Shortage," 2026
- 3.GE Vernova's total annual gas turbine production runs on the order of 20 GW worldwide, for every customer on the planet. Yahoo Finance, "The Gas Turbine Shortage Just Became AI's Biggest Constraint," August 2026
- 4.Exelon cut its "high probability" data center load from roughly 18 GW to about 11 GW in a single quarter, and Texas ordered an audit of the ERCOT interconnection queue after requests swelled to roughly 474 GW, about 90% of it data centers. Utility Dive, Exelon earnings coverage, July 2026; Utility Dive, "Texas hits pause on data center interconnections," August 2026