Subscribe
PROLONG, an AI “time machine”, uses 13,000 virtual worlds and real-world data from 200+ countries to forecast wind and solar growth.
It projects onshore wind at 25% and solar at 20% of global power by 2050, making the COP28 tripling pledge highly ambitious.
This is enough for a 2°C pathway but falls short of the 1.5°C goal.
Wind and solar power have expanded far faster than earlier projections, but forecasting their future growth remains complex. A new study published in Nature Energy introduces an AI-driven model that improves predictions of global wind and solar expansion, projecting strong growth by mid-century but not enough to meet the most ambitious climate targets.
Researchers from Chalmers University of Technology developed an AI-based computational time machine, called PROLONG, that analyses historical national trends across more than 200 countries. Unlike conventional approaches, the model captures real-world patterns of uneven growth, including sudden policy-driven surges.
The model estimates that by 2050, onshore wind could generate around 25 per cent of global electricity, while solar may contribute about 20 per cent — figures aligned with a 2°C pathway but falling short of what is required to meet the 1.5°C climate goal.
“Existing models are very good at identifying what needs to happen to reach climate targets, but they can’t tell us which developments are most likely. That’s the gap we wanted to fill,” said Jessica Jewell, professor at Chalmers University of Technology.
The study finds that renewable energy expansion does not follow the smooth, S-shaped trajectories assumed in many models. Instead, countries experience long periods of steady growth punctuated by sharp accelerations, often triggered by policy shifts.
“Most models assume a smooth S-shaped growth curve, but that’s not how it actually looks in the real world. Growth often comes in bursts, and if you ignore that, you can misjudge how fast technologies will expand,” said Avi Jakhmola, PhD student at Chalmers and first author of the study.
These dynamics reflect competing forces: Falling costs and technological learning on one side, and rising constraints such as land availability, grid integration challenges and public opposition on the other.
To improve forecasts, the researchers built a model trained on 13,000 simulated “virtual worlds”, each representing different pathways for wind and solar growth — from rapid acceleration to slow expansion.
A machine learning algorithm then learnt to connect early national trends with likely global outcomes.
“When we apply the model to real-world data, it can tell us what is the most probable outcome for the future, given what we have seen so far and given all the virtual worlds it has seen,” said Jakhmola.
The model also demonstrated strong reliability in back-testing. Using only data up to 2015, it accurately predicted actual global deployment trends in the years that followed.
The findings put global policy targets into perspective. The pledge made at COP28 to triple renewable energy capacity by 2030 fell near the 95th percentile of projected outcomes, meaning it would require unusually rapid growth rarely observed historically.
“The tripling of renewables pledge is not impossible, but it would require everything to go extremely well in all countries,” said Jewell.
The study also highlights the urgency of immediate action. Achieving a 1.5°C-compatible trajectory would require growth rates comparable to aggressive current policies, such as those seen in Europe and planned in India — but delays would sharply increase the challenge.
“If we start now, the required growth rates are demanding but not unprecedented… But if we delay until 2030, the acceleration needed becomes much steeper and much more abrupt. The window for ramping up closes quickly,” said Jakhmola.
The researchers argued that their approach filled a key gap between scenario-based models — which outline what is needed — and real-world projections of what is likely.
“It’s long been a joke how bad technology forecasts are. But if you’re a decision maker, trying to figure out how hard to push for change, you need a realistic baseline. Our study is the first step towards developing such a realistic view of the future,” said Jewell.
Beyond wind and solar, the team says the method could be applied to other low-carbon technologies, helping policymakers better align ambition with achievable pathways.