The Grid-Equipment Order Book: Who Supplies AI’s Power Build-Out

Order a heavy-duty gas turbine from GE Vernova today, and it will not arrive until 2031. That single line explains the artificial intelligence build-out better than any data-center ribbon-cutting. The constraint stopped being chips a while ago. It now sits in the physical equipment that turns fuel into electricity: turbines, transformers, switchgear, and the factories that build them.

The queue is the story.

Three years ago the race for artificial intelligence ran through graphics processors. Now it runs through order books. Gas turbines are sold out past 2028, and manufacturers are taking reservations for deliveries five years out. For a principal weighing where value in this cycle actually accrues, the equipment layer is doing more work than the model layer.

Gas is winning the near-term race for one unglamorous reason: it is dispatchable. A solar farm or a battery cannot yet guarantee round-the-clock output at the scale a training campus needs, and grid interconnection queues stretch for years, which pushes developers toward generation built behind the meter. Aeroderivative turbines, the lighter and faster siblings of the heavy-duty machines, can be installed in months rather than years. Henry Hub spot gas traded near $2.79 per million British thermal units in mid-August 2026, a price that keeps on-site generation economical against grid power in most US markets.

The Order Book Is the Bottleneck

The second-quarter 2026 results, reported on July 22, put the number plainly. Total order backlog across the company reached $176 billion, up from $129 billion a year earlier. The Power segment booked $16.7 billion in orders, a 134% organic jump. The gas-turbine backlog and slot reservations climbed to 116 GW from 100 GW three months earlier, and management expects at least 125 GW under contract by year-end.

Those figures reframe what "powering AI" means. A hyperscaler announcing a gigawatt campus is announcing demand. The company that can promise a delivery date owns the negotiation.

The backlog splits into two parts worth separating. Roughly 53 GW is firm equipment backlog, and about 63 GW is paid slot reservations, the queue where a customer holds a place for a machine years out. Reservations are the leading indicator. They convert to firm orders as projects reach financial close, and it converted 10 GW of them in the second quarter alone.

Then there is the quieter number. Electrification, the segment that makes transformers, switchgear, and substations, reached a $41 billion backlog, up 69% year over year. Data-center electrification orders in the first half of 2026 passed $5 billion, more than the whole of 2025. Transformers are the component that most reliably slows a project. Large units take two to five years to build, and their lead times have stretched alongside the turbines.

The International Energy Agency framed the scale in its World Energy Investment report. If data centers were a country, the agency said, they would have been the second-largest destination for gas turbines ordered between early 2025 and early 2026. New orders for US data-center power outpaced the total booked by any other single nation.

Two Portfolios, Two Directions

Its segments are moving in different directions. Wind equipment orders fell 40% year over year, pressed by soft onshore demand in the United States, while gas and electrification carried the quarter. Siemens Energy shows a similar split, with the gas-services division as the engine and the grid business as the beneficiary.

Pricing followed.

Turbine prices for plants entering service in the early 2030s have risen well above where they sat a few years ago, and the manufacturers are adding capacity to chase the demand. It has brought roughly 30 additional medium gas-turbine manufacturing units online since 2025 and plans to lift transformer capacity by 50% by 2030. Baker Hughes' industrial and energy technology (IET) backlog reached a record $37.1 billion, built on both liquefied natural gas (LNG) equipment and data-center power demand.

The historical caution is worth keeping in view. In the 2010s, a wave of gas-turbine orders collided with weak electricity demand and cheap renewables, leaving manufacturers with idle capacity and billions in charges. GE's purchase of Alstom's power business became one of the costliest bets in the sector's history. The difference this time is the buyer. Data centers are a concentrated, contract-backed customer with a mandate to secure power quickly, signing long-term agreements rather than waiting on merchant markets.

The New Layer: Modular and Hypermobile Power

The order book is no longer only about the four large manufacturers. Integrators are packaging turbines into fast-deployable power blocks aimed at the window before grid capacity arrives.

In August, Dynamis Power Solutions ordered 76 NovaLT16 industrial gas turbines from Baker Hughes, about 1.3 GW, for a hypermobile platform it calls DT17. Days later, Babcock & Wilcox engaged Siemens Energy on 20 steam turbine-generator sets totaling 1 GW for its FastPower program, aimed at data-center customers. Both products target the same gap: a campus that cannot wait three years for a grid connection.

The larger players are following the same logic with long-term commitments. NextEra Energy struck a framework agreement with GE Vernova to co-develop multiple gigawatts of gas-fired generation, renewables, and storage for large loads, alongside a plan to restart the Duane Arnold nuclear plant in Iowa. Chevron and GE Vernova are targeting 4 GW of data-center power by 2027. Crusoe runs LM2500XPRESS aeroderivative units at its AI campus.

A secondary market is forming around scarcity itself. Suppliers now hold inventory of turbines, transformers, and high-voltage circuit breakers to cut the lead times that grid interconnection schedules impose, selling time-to-power as a product.

What the Backlog Prices In

The bull case is visibility. Backlogs stretching to 2031 convert a cyclical manufacturing business into something closer to an annuity, and the aftermarket follows the equipment: every turbine shipped carries decades of service revenue. Manufacturers are adding capacity, the clearest sign they expect the demand to persist rather than spike.

The risks are just as plain. Order books built on hyperscaler capital spending inherit hyperscaler risk. If AI infrastructure budgets pause, slot reservations can be cancelled or deferred, and a backlog can shrink as fast as it grew. The wind segment already shows what a demand air pocket looks like. A gas-price spike or a permitting reform that shortens interconnection queues would each soften the order flow.

As we wrote in September, SB Energy filed to raise $5-7 billion against a $439 billion backlog and 8.8 GW of contracted AI data-center capacity, none of it operating. The contracts exist. The energized megawatts do not. That gap is exactly why the equipment queue, not the project pipeline, is the number to watch.

Signals to Track