AI’s Newest Bottleneck: The Gas Turbine Shortage
admin 
**The Gas Turbine Shortage Just Became AI’s Biggest Constraint**
If you order a heavy-duty gas turbine from GE Vernova today, it will not arrive until 2031. This reality, confirmed during the company’s July earnings call, highlights a massive bottleneck that threatens to derail the ambitious power plans of AI data centers worldwide.
According to projections from Goldman Sachs, U.S. data center power demand is on track to rise from 31 gigawatts (GW) in 2025 to 41 GW in 2026, reaching 66 GW by 2027. This represents an annual capacity addition of 36.3 GW in 2027 alone—a figure that exceeds the total manufacturing capacity of the world’s leading turbine producers combined.
### The Industry is Booked Solid
The three major global gas turbine manufacturers are facing unprecedented backlogs:
* **GE Vernova** finished the second quarter of 2026 with 116 GW of gas power equipment backlog and slot reservation agreements. The company is currently booking delivery slots for 2031 and plans to scale its annual production from 20 GW to 30 GW by 2030.
* **Siemens Energy** reported a gas turbine backlog of nearly 70 GW, with lead times stretching to three years or more.
* **Mitsubishi Heavy Industries** reported a 35 GW backlog for its large-frame turbines, with new orders scheduled for delivery between 2028 and 2030.
Combined, global orders reached a record 38 GW in the second quarter, up 71% year-over-year, with the U.S. accounting for half of that demand. According to Wood Mackenzie, global manufacturing capacity stands at just 60 to 70 GW per year, compared to approximately 110 GW of annual orders.
### A Supply Chain Bottleneck
The mismatch between AI’s power needs and the industrial supply chain is stark. Turbines required to power data centers in 2027 had to be ordered in 2023 or 2024. Any equipment ordered today will not arrive until the next decade.
Furthermore, this supply chain is notoriously difficult to scale. Production relies on a small number of specialized foundries capable of producing high-temperature hot-section castings. Additionally, there is a severe shortage of the skilled welders, machinists, and specialized contractors required to assemble and install these massive machines. As a result, the average lead time for a new combined-cycle power plant has stretched from 3.5 years in 2023 to roughly 5 to 7 years today.
### Skyrocketing Costs and Grid Strain
This equipment deficit is driving power prices and construction costs to record highs:
* **PJM Interconnection’s** recent capacity auction cleared at its price cap of $325 per megawatt-day for the third consecutive year. Despite the record-high pricing, the auction secured 138,318 MW of capacity—leaving the grid 6,831 MW short of its reliability requirement while attracting only 525 MW of new generation.
* **Plant Construction Costs** have surged. BloombergNEF reports that the average cost of a combined-cycle project reached $2,157 per kilowatt (kW) last year, up from under $1,500 in 2023. Wood Mackenzie expects turbine-only prices to hit $600 per kW by late 2027, a nearly 200% increase from 2019 levels.
Notably, these rising costs are entirely driven by manufacturing bottlenecks, not fuel prices. Henry Hub spot natural gas continues to trade at historically unremarkable levels.
### Speculative Queues and Regulatory Pushback
To protect themselves from potential market corrections, turbine manufacturers now require heavy down payments for “slot reservations.” This shifts the financial risk to developers, allowing manufacturers to collect non-refundable deposits even if projects are ultimately canceled.
In response to the gridlock, utilities and state regulators are beginning to scrub speculative projects from their pipelines:
* **Exelon** recently cut its “high probability” data center load forecast by nearly 40%, dropping it from 18 GW to 11 GW to weed out speculative requests.
* **Texas** governor Greg Abbott ordered an audit of all data centers in the ERCOT queue after connection requests swelled to 474 GW—more than five times the state’s peak demand. ERCOT has suspended its fast-track approval process for large loads, putting nearly 50 GW of planned capacity at risk of delay.
* **New York** also halted new approvals for data center interconnections earlier this summer.
### The Path Forward
Because the turbine bottleneck cannot be resolved quickly on the factory floor, the power gap in the near term will have to be met by alternative solutions. Developers are increasingly turning to reciprocating natural gas engines from manufacturers like Caterpillar and Wärtsilä, which can be delivered faster than heavy-duty turbines. Other workarounds include fuel cells, running existing fossil-fuel plants at higher capacities, and delaying the retirement of coal-fired units.
Ultimately, the constraint on the AI revolution is no longer just chips or software, but the physical infrastructure and regulatory approvals required to keep the lights on.
