The combined-cycle gas-turbine (CCGT) fleet AI data centres now lean on is middle-aged: across 2,251 operating CCGT plants with a known commissioning date, the median is 18 years old and 955 (42%, ~534 GW) are over 20 years old. In the EU, 267 of 511 CCGT plants pre-date 2006. With new gas turbines carrying 5-7 year lead times, the 2026-2030 AI load runs largely on this existing, ageing fleet.
Data-centre electricity demand is the fastest-growing industrial load of the decade and is met, near-term, mostly by gas. But new capacity is supply-constrained: heavy-duty gas turbines carry 5-7 year lead times and OEM backlogs above annual build capacity. So the AI load of 2026-2030 runs on the fleet already standing — and it is not new.
PowerAtlas CCGT/HRSG fleet inventory (per-plant commissioning year), built on WRI GPPD + Global Energy Monitor. CC BY 4.0.
| Country | CCGT plants | >20 yrs |
|---|---|---|
| United States | 501 | 65% |
| China | 236 | 12% |
| Germany | 109 | 51% |
| Italy | 93 | 52% |
| Spain | 77 | 56% |
| Mexico | 73 | 45% |
| United Kingdom | 69 | 51% |
| India | 65 | 55% |
| Thailand | 62 | 21% |
| Japan | 59 | 46% |
| South Korea | 58 | 28% |
| Türkiye | 54 | 33% |
EU subset: 267 of 511 plants over 20 years old.
| Plant | Country | Commissioned | Age | MW |
|---|---|---|---|---|
| Queen Elizabeth generating station | Canada | 1959 | 67 | 486 |
| Emsland power station | Germany | 1973 | 53 | 1,837 |
| Franken 1 power station | Germany | 1973 | 53 | 835 |
| Gersteinwerk power station | Germany | 1973 | 53 | 820 |
| Marbach power station | Germany | 1974 | 52 | 265 |
| Brunot Island power station | United States | 1974 | 52 | 340 |
| Comanche (OK) power station | United States | 1974 | 52 | 290 |
| Santan power station | United States | 1974 | 52 | 1,328 |
| T H Wharton power station | United States | 1974 | 52 | 664 |
| SWB Mittelsbüren power station | Germany | 1975 | 51 | 238 |
| Newman power station | United States | 1975 | 51 | 597 |
| Gómez Palacio power station | Mexico | 1976 | 50 | 239 |
≥200 MW, by age. Full data: PowerAtlas CCGT/HRSG fleet.
The age problem matters because the alternatives are slow too. Lead time — order to power-on — is now the binding variable for AI data centres:
| Power source | Typical lead time | Fit for 24/7 AI load |
|---|---|---|
| Solar + battery storage | 1-2 yrs | Cheap & fast, but intermittent — needs large storage for round-the-clock |
| Onshore wind | 2-3 yrs | Variable; siting & grid-connection limited |
| Gas turbine (CCGT) | 5-7 yrs | Dispatchable baseload — but emits CO₂ and the fleet is ageing (above) |
| Small modular reactor (SMR) | ~5-7 yrs (first units) | Carbon-free 24/7 — not yet commercial at scale |
| Large nuclear (new build) | 10-15 yrs | Carbon-free baseload — slowest to build |
Indicative industry lead times, 2026. This is precisely why near-term AI load leans on the existing (ageing) gas fleet and nuclear restarts rather than new build.
Every gas turbine, HRSG and steam line in this fleet — and every kW of data-centre waste heat — is a hot surface. Inzonex modular removable insulation cuts surface heat loss up to 96% while keeping flanges accessible for maintenance.
See Inzonex modular insulation →Across 2,251 operating CCGT plants with a known start date, the median is 18 years old and 955 (42%) are over 20 years old (~534 GW).
Heavy-duty gas turbines carry 5-7 year lead times as of 2026 and the OEM order backlog (GE Vernova, Siemens Energy, Mitsubishi) exceeds annual build capacity, so near-term AI load relies on the existing fleet.
Age alone does not set emissions — output depends on turbine model, maintenance and duty cycle, for which public per-plant heat-rate data is sparse. This page reports verified commissioning age only, not modelled efficiency.
Per-plant data: PowerAtlas CCGT/HRSG fleet inventory (2,251 operating plants with commissioning year), from WRI GPPD + Global Energy Monitor, CC BY 4.0. Lead-time/backlog from public OEM reporting 2025-2026. No per-plant efficiency estimated (no public heat-rate data). Author: Dmytro Aheiev (ORCID 0009-0001-5512-0291).
Externally-sourced figures are attributed and link to primary sources; deal terms are public company/trade-press announcements. Author: Dmytro Aheiev (ORCID 0009-0001-5512-0291).