S06.5 · Energy, Power & Climate

Conventional Power Generation

Electricity generated from thermal and nuclear sources for grid supply, now being re-contracted directly to hyperscalers to meet AI datacenter demand.

S06.5

What is on this page. Market structure, and how AI is reshaping this segment. Ownership, buyer universes, transaction comparables and deal-timing analysis are maintained privately by El Dorado Capital and are not published.

Conventional power generation — electricity from thermal and nuclear sources for grid supply — covers roughly 5,500-6,000 GW of installed capacity globally in 2025 (IEA Electricity 2026), with generation revenue estimated at $1.5-2T. Capacity growth is slowing across the OECD and continuing in Asia, particularly Chinese and Indian coal and gas, while nuclear sees renewed order activity including small modular reactors. Alone in this sector, the AI story here is demand, not cost: datacenter load growth has utilities and independent power producers re-contracting existing and new thermal and nuclear capacity directly to hyperscalers. Nothing about how the plants run is changing. Who they sell to is.

Market structure

China holds about half of the global coal fleet; gas generation concentrates in the US and Middle East; nuclear in France, the US and China. The market split matters more than the geography. Regulated cost-of-service markets — most of the US, the EU and Asia — and deregulated wholesale markets — ERCOT, PJM, the UK, EU wholesale and Australia's NEM — build revenue on genuinely different structures, and that distinction governs how a given asset earns its return. Ownership is fragmented across utilities and independent power producers, with the top 10 global generators holding under 15% combined share.

Generators buy fuel from upstream and midstream and sell power to utilities, retailers and grid operators. The economics stack high capital intensity, a mix of regulated rate-base and merchant/wholesale exposure, fuel pass-through, carbon-regulation exposure and asset lives of 30-60 years, with permitting a heavy gate — especially for nuclear. That asset life is central to how datacenter power deals are structured. A hyperscaler contracting for output from an existing thermal or nuclear plant is buying dispatchable, already-permitted capacity rather than waiting years for new-build — which is precisely why existing conventional capacity, not new renewable build, has become the fastest route to serving incremental AI load.

How AI is reshaping this segment

Carbon-capture retrofits defend existing asset life under tightening emissions rules. Dedicating gas peaker fleets and nuclear output to datacenters does something rarer in a mature segment: it captures an entirely new demand pool.

The operative dynamic is demand-side, and it is already priced. Hyperscaler power deals with generators such as Constellation and Vistra are signed and operating, re-pricing access to existing thermal and nuclear capacity rather than changing how that capacity is generated or dispatched. For the generator holding the asset this is new, incremental contracted revenue — net-new budget from the buyer, not a reallocation of existing utility spend — and it concentrates on operators with dispatchable, already-permitted capacity near datacenter clusters rather than spreading evenly across the generation fleet.