AI’s Hidden Power Crisis: Why Gas Turbines May Stall the Next Wave of Data Centers

Order a heavy-duty gas turbine from GE Vernova today. It won’t arrive until 2031. The statement lands with force. It comes straight from the company’s July earnings call. And it sits quietly beneath every hyperscale data center announcement of the past two years.

Data center operators tout gigawatts of new capacity. Tech executives promise AI breakthroughs. Yet the machines that turn natural gas into the steady electrons those facilities demand face backlogs stretching deep into the next decade. Goldman Sachs laid out the numbers in May. U.S. data center power demand rises from 31 gigawatts in 2025 to 41 GW this year and 66 GW in 2027. Goldman Sachs report. Capacity additions jump from 8.5 GW realized last year to 13.6 GW scheduled for 2026 and 36.3 GW in 2027. Data centers would then claim 8.5 percent of total U.S. peak summer demand. Up from 4.1 percent today.

The equipment lags far behind. GE Vernova closed its second quarter with 116 GW of gas power equipment backlog and slot reservation agreements. Up from 100 GW three months earlier and 83 GW at year-end 2025. The company expects at least 125 GW under contract by December. CEO Scott Strazik told analysts it has begun taking reservations for 2031 delivery. It should stand more than halfway contracted for that year by the end of 2026. Production targets sit at roughly 20 GW annualized this quarter. Then 24 GW by 2028. And a push toward 30 GW by 2030. OilPrice.com.

Siemens Energy finished its fiscal third quarter on June 30 with a 69 GW gas turbine backlog. It booked 15 GW and shipped six. Lead times run three years or longer. CEO Christian Bruch noted the addressable market could hit 120 GW a year. With roughly half in the United States. Mitsubishi Heavy Industries reported a 35 GW large-frame backlog on Aug. 6. Up from 23 GW a year earlier. CFO Hiroshi Nishio said orders booked that quarter target delivery between 2028 and 2030. The company stays selective. “being selective in the projects we contract.”

These figures don’t line up perfectly. GE Vernova’s 116 GW includes only 53 GW in firm equipment backlog. The rest sits in paid slot reservations. Siemens counts firm backlog alone. Mitsubishi focuses on large-frame units. Still the combined picture exceeds 220 GW. Global manufacturing capacity hovers between 60 GW and 70 GW per year. Demand has already surpassed that level. Crypto Briefing.

Recent updates reinforce the strain. GE Vernova’s power segment booked $16.7 billion in orders in the second quarter of 2026. Up 134 percent organically. Data center-driven electrification orders exceeded $5 billion year-to-date. Potential revenue per GW has risen two to three times with new technologies like solid-state transformers. Scott Strazik repeated the production goals. Twenty GW in the third quarter of 2026. Twenty-four GW in 2028. Actions aimed at 30 GW in 2030. Much of the added capacity fits inside existing factory footprints. Turbomachinery Magazine.

Baker Hughes has moved aggressively into the space. Its BRUSH generators now pair with 42-MW Superpower gas turbines for off-grid AI data centers. Twenty20 Energy contracted for 10 Frame-5 turbines and generators in Georgia and Texas. Initial capacity reaches 250 MW. Deliveries begin in 2027. Boom Supersonic holds a $1.2 billion backlog for its Superpower turbines. Crusoe ordered 29 units. Lorenzo Simonelli, Baker Hughes chairman and CEO, described a decade-long cycle driven by turbomachinery, LNG and AI data center power. The company booked $1 billion in data center orders in the first quarter of 2026 alone. It raised its 2025-2027 targets. Turbomachinery Magazine.

Supply chain limits compound the problem. Skilled labor shortages slow factory ramps. Specialized materials face global constraints. Complex components require precision manufacturing that cannot scale overnight. Manufacturers remember past boom-and-bust cycles. They expand cautiously. Even optimistic plans from the big three add only about 25 percent more output by the end of the decade. Demand grows faster. PJM Interconnection’s July 2026 capacity auction fell 6,831 MW short of its reliability target. It cleared at the price cap for a third straight year. New generation added just 525 MW. OilPrice.com.

Costs have climbed sharply. Capital expenses for combined-cycle gas plants reached $2,157 per kilowatt in 2025. Up from under $1,500 per kilowatt in 2023. A 44 percent jump in two years according to some analyses. Others report costs doubling to $2,400 per kilowatt. Construction timelines stretch 23 percent longer. Reservation fees appear. One Kentucky utility paid GE Vernova $25 million simply to hold a 2030 delivery slot. Prices for turbines themselves carry premiums for faster positions. Turbomachinery Magazine.

Developers adapt. Some turn to medium industrial turbines and aeroderivatives for quicker delivery. Solar Turbines, part of Caterpillar, saw sales rise 23 percent and plans to expand capacity 2.5 times by 2030. Reciprocating engines from suppliers like Wartsila deliver in 12 to 18 months in some cases. Yet they carry higher emissions and lower efficiency. Off-grid modular blocks of 150 to 300 MW gain traction. They bypass grid interconnection queues that can stretch years. But they still need turbines or similar rotating equipment. Reuters.

The scramble reaches beyond the United States. Alberta’s plan to attract $100 billion in data centers collides with the same global shortage. Wait times there stretch five years or more. Project costs rise. Asian utilities find themselves pushed back in line as U.S. data center demand claims priority. Some Vietnamese developers already warn that gas plants targeted for operation within three years will miss deadlines. Coal plants gain fresh consideration in certain markets despite emissions goals. Bloomberg analysts calculate more than $400 billion in planned gas-fired projects through 2030 now risk delay or cancellation. The Narwhal.

Grid operators sound alarms. Data centers could face curtailment risks during peak stress. One regional transmission organization has signaled new data centers might rank first in line for load shedding. That prospect clashes with the always-on expectations of AI training clusters. Hyperscalers respond with direct power purchase agreements. Some pursue nuclear restarts or small modular reactors. Others explore fuel cells and on-site generation. None eliminates the immediate turbine gap. Bloomberg.

Industry voices stay measured. Klaus Brun and Rainer Kurz examined the 10-year outlook. They note 2024 posted the highest number of gas turbine units ordered since 2002. Orders rose 32 percent in power terms from 2023. Yet they question whether this marks a permanent shift or another cycle. Delivery times have doubled. Some buyers now wait seven years. The surge ties to both AI and the need for dispatchable power to balance renewables. Turbomachinery Magazine.

Richard Reisig, managing director of global investments and asset management at Siemens Energy, offered a blunt assessment. About 25 percent of the company’s current gas turbine order book ties directly to data centers. A year ago that share was negligible. Scaling stays constrained by labor, materials and component complexity. BloombergNEF reports planned U.S. gas capacity has grown nearly sixfold while manufacturing capacity lags. Average project costs for gas plants climbed 66 percent in two years. Turbomachinery Magazine.

The bottleneck reshapes priorities. Tech companies once assumed power would appear when needed. Now they compete for physical hardware produced in a handful of factories worldwide. GE Vernova, Siemens Energy and Mitsubishi Heavy Industries control more than 75 percent of large-frame supply. Their caution reflects history. Over-ordering in the early 2000s left scars. Today’s restraint keeps lead times long. And it forces difficult choices on everyone downstream.

Short-term relief looks limited. New manufacturing lines take years to qualify. Workforce training cannot accelerate dramatically. Material supply chains for high-temperature alloys and precision blades remain tight. Even if all announced expansions hit targets, annual output may reach only 80 to 90 GW by 2030. Projections show annual orders exceeding 60 GW this year and staying above global production thereafter. The math doesn’t close without trade-offs.

Hyperscalers may pay more. They may accept slower buildouts. Some will shift to smaller, faster-deploying equipment even if it raises operating costs. Others explore behind-the-meter solutions that reduce grid dependence. The common thread remains clear. The turbines represent a fixed, slow-moving constraint in an industry that moves at digital speed. Until production catches up or demand moderates, this shortage dictates the pace of AI infrastructure growth. The announcements will keep coming. The electrons, however, must wait their turn.