Surprising trends in global electricity generation

Start with the measurement choice, because it is doing more work than it looks like. Ember reports generation rather than capacity. Capacity is a good leading indicator of the next six to twelve months, but generation is what is actually happening on the ground, and the gap between the two turns out to be where most of the surprises live. On that measure solar added 636 terawatt-hours in 2025, roughly twice the United Kingdom’s annual electricity demand. The number that matters is not the absolute record but the rate: growth rates are supposed to fall as a market matures, and solar’s did not. It grew 30%, the highest in eight years, against a ten-year average of 27%, which puts 2025 inside a decade of consistent exponential growth rather than above it. Fulghum’s historical frame is that no source has ever climbed from 100 terawatt-hours to 2,000 as fast as solar, with wind a few years behind and nuclear in the 1970s and 80s the only comparable trajectory. That last comparison is a warning as much as a compliment, because nuclear levelled off almost immediately afterwards and added little net generation globally in the decades that followed. The live question is when solar’s S-curve turns, because the turn’s timing determines whether the world reaches 30, 40 or 50% solar or stalls near the 8 to 9% where it sits now.

The fossil decline follows from simple arithmetic, which is what makes it fragile in the right way. Fossil generation falls in any year when the increase in clean generation exceeds the increase in demand, and 2025 is simply the year clean growth came out a little ahead: solar’s 636 terawatt-hours plus wind’s 200-plus, with hydro flat and a small nuclear increase, against 849 terawatt-hours of demand growth. The country-level detail is where the headline gets slippery. In China the mechanism is structural. Clean growth is meeting rapidly rising demand, fossil generation has now been flat for a couple of years, and Fulghum says the country arrived at that point earlier than most people anticipated. In India the same-looking outcome had a different cause: a mild monsoon season kept temperatures down, and Indian demand is heavily pegged to cooling, so demand growth was unusually weak. Pair weak demand growth with a record renewables increase and fossil generation falls without the underlying trend being finished. Fulghum expects India to need a few more years, closer to 2030, before it posts fossil declines in a normal demand year. Both countries fell; only one of them has crossed over.

China’s coal is the most counterintuitive part, and it is where generation and capacity point in opposite directions. Coal capacity is still growing, which produces the headlines about China betting on coal that Fulghum calls misleading. The plants being built now trace back to generation shortfalls in 2020 and especially 2021 and the policy reaction that followed, and China can turn a coal plant around in three or four years where other countries cannot, so a pipeline persists. The generation data says something else is happening to those plants. Coal’s capacity factor has fallen consistently, and actual policy rather than commitments is lowering the minimum output level coal units are allowed to run at, so they can throttle further down at midday and make room for solar. Kann proposes the mechanism, that new coal plants are effectively becoming shoulder and peaking resources rather than baseload, and Fulghum confirms it directly. What makes this rational rather than merely wasteful is the cost asymmetry: every unit of coal generation burns fuel that costs money, while solar’s cost is capacity already paid for. The catch is curtailment, which is already high and which Fulghum believes the official numbers understate, his evidence being that 2026 solar generation is growing more slowly than capacity additions imply. Dispatch rules written to favour coal take multi-year market reform to unwind. He reads the shortfall as a backlog rather than a loss, since integrating that capacity later unlocks generation without anyone building anything new.

Batteries get the most useful new metric in the report, and it reverses an intuition. Ember compares daily added solar generation to added battery energy capacity: 636 terawatt-hours of new solar is a bit under 2 terawatt-hours a day, and 2025’s roughly 250 gigawatt-hours of new batteries can therefore time-shift about 14% of the new solar, assuming one charge and discharge cycle per day. The obvious reading is that 14% is a long way from the 100% you would want. But you do not want 100%. About 50% is the point at which batteries can take the entire midday peak away, because the morning and afternoon solar should stay where it is, in hours the grid still has room for. Chile and Australia are already above that line, which is why Australian and Californian solar growth now shows up in the shoulder hours and the evening rather than at noon, and why Fulghum argues cheap batteries make solar effectively dispatchable and dissolve the penetration thresholds that used to bound it. Kann draws the investable inference, that batteries should grow two to three times faster than solar until the ratio reaches roughly 50%. Fulghum’s complication is that the path will not be linear, because battery deployment is gated on revenue models and offtake rather than on cost or technology, so a market can sit near zero with everything else in place and then take off in a burst. Both point to ERCOT as the cautionary case, where an early boom helped compress spreads and left merchant batteries underperforming, which makes the next wave harder to finance even as solar and wind keep building.