On April 27, 2026, TechCrunch reported on new BloombergNEF (BNEF) figures showing that the cost of building a new combined-cycle gas turbine (CCGT) plant in the U.S. has risen 66% over the past two years. Per-kilowatt construction costs climbed from under $1,500/kW in 2023 to $2,157/kW in 2025, while build times stretched 23% longer. The primary driver has one name: AI data center demand.
These numbers mark the moment the AI buildout officially became a power problem rather than a chip problem. When the electricity appetite of training and inference grows faster than grids and plants can be expanded, power itself becomes a pricing factor for compute. For any team planning large-scale AI deployment beyond 2026, this is not a forecast — it is the reality of the next two procurement cycles.
Where the 66% Comes From: $1,500 to $2,157 per kW
BNEF’s benchmark is construction cost per kilowatt of generating capacity. In 2023, a new CCGT could be built for under $1,500/kW; by 2025 the figure hit $2,157/kW, a cumulative 66% rise. Time is the second shock: completing a new facility now takes 23% longer. Against a backdrop of U.S. natural gas prices staying low, the increase comes almost entirely from equipment and engineering scarcity, not fuel.
In project terms: a 1 GW-class gas plant now carries roughly $660 million more in construction cost than a two-year-old budget assumed, and it waits in a longer queue before ground breaks. For AI loads that need megawatts quickly and predictably, those three conditions no longer coexist.
Gas Turbines: The Hardest Bottleneck in the Chain
The core of the cost spike is the heavy-duty gas turbine. Wood Mackenzie’s April 1 report estimated turbine prices will reach $600/kW by the end of 2027, up 195% from 2019, and turbines account for roughly 20–30% of a combined-cycle project’s cost — more for simple-cycle plants. The supply-demand gap is stark: at the end of 2025, global order books stood at about 110 GW against manufacturing capacity of only 60–70 GW. Order books are sold through 2027, with lead times stretching to six years.
Manufacturers are expanding, but slowly. GE Vernova is spending over $160 million to lift large-frame turbine output from about 50 units a year to 70–80 by late 2026. Siemens Energy has moved key facilities to 24/7 operations alongside a $1 billion U.S. investment program. Mitsubishi Heavy Industries plans to double capacity by 2028. The binding constraints are components and people: hot-section parts such as single-crystal blades come from only a handful of global suppliers, and specialized labor is short.
The scale of demand is visible in a single project: SB Energy’s Portsmouth Powered Land scheme — 9.2 GW and $33 billion, announced in February 2026 — needs 24 to 30 heavy-duty turbines for its first phase alone. Wood Mackenzie expects 63 GW of U.S. gas additions between 2026 and 2030, with orders peaking in 2026.
The Demand Side: From 40 GW to 106 GW
BNEF’s data center forecast is just as steep: from roughly 40 GW today to 106 GW by 2035, a 2.7x increase. Facility shape is changing too — only about 10% of data centers are 50 MW or larger today, but over the next decade the average facility will exceed 100 MW. Wood Mackenzie separately projects data center power consumption growing 96% between 2026 and 2031, the fastest-growing new load on the U.S. grid.
That explains why cost and schedule broke simultaneously. The load arrives in huge, lumpy increments — a single hyperscaler campus can absorb a full plant’s output — while the supply chain for generation expands on a cadence measured in years, sometimes decades.
Two Routes: Gas Versus Renewables Plus Storage
Hyperscalers are splitting into two postures. Microsoft and Meta are racing to build their own gas plants, pairing generation directly with data centers, and the Trump administration has publicly urged operators to “bring their own power” and lean less on the public grid. That route carries political cost: utilities typically pass new generation costs to ordinary ratepayers, which has already fueled a backlash that makes “data center” a loaded word in local permitting fights.
Google is betting the other way: renewables plus long-duration storage. Its playbook includes Form Energy’s iron-air batteries, which discharge for over 100 hours — far beyond the 4-hour class of lithium — pairing solar-plus-storage against gas. With solar and battery costs still falling while turbine prices surge in the opposite direction, the crossover point between the two routes is arriving earlier than most models assumed.
What It Means for Infrastructure and Product Teams
Three practical effects. First, power acquisition is now a first-class constraint on AI deployment: siting, latency, and go-live planning must account for turbine lead times and grid interconnection queues, not just GPU supply. Second, cost structures will propagate — whether you build plants, sign PPAs, or rent cloud GPUs, rising power prices and community opposition are real project risks, and where your inference workloads live determines your unit economics. Third, the race between clean firm power with long-duration storage and gas will set the carbon and cost curve of new compute for the next five years. Disclosing AI energy and emissions data is shifting from a marketing option to a compliance and community-relations requirement.
Sources
- Data center demand drives 66% surge in natural gas power plant costs — TechCrunch
- Cost to Build Natural Gas-Fired Power Plant Surges 66%, BNEF Says — Bloomberg
- Gas turbine prices soar 195% as market faces supply-demand crisis — Wood Mackenzie
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.