It’s tricky to make any predictions about the cost of a technology that doesn’t exist yet. But when there’s billions of dollars of public and private funding on the line, it’s worth considering what assumptions we’re making about our future energy mix and its cost.

One crucial measure is a metric called experience rate—the percentage by which an energy technology’s cost declines every time capacity doubles. A higher figure means a quicker price drop and better economic gains with scaling.

Historically, the experience rate is 12% for onshore wind power, 20% for lithium-ion batteries, and 23% for solar modules. Other energy technologies haven’t gotten cheap quite as quickly—fission is at just 2%.

In the new study, published in Nature Energy, researchers aimed to improve predictions of fusion’s future price by estimating the technology’s experience rate. The team looked at three key characteristics that can correlate with experience rate: unit size, design complexity, and the need for customization. The larger and more complex a technology is, and/or the more it needs to be customized for different use cases, the lower the experience rate.

The researchers interviewed fusion experts, including public-sector researchers and those working at companies in the private sector. They had the experts evaluate fusion power plants on those characteristics and used that info to predict the experience rate. (One note here: The study focused only on magnetic confinement and laser inertial confinement, two of the leading fusion approaches, which together receive the vast majority of funding today. Other approaches could come with different cost benefits.)

Fusion plants will likely be relatively large, similar to other types of facilities (like coal and fission power plants) that rely on generating heat. They will probably need less customization than fission plants—largely because regulations and safety considerations should be simpler—but more than technologies like solar panels. And as for complexity, “there was almost unanimous agreement that fusion is incredibly complex,” says Lingxi Tang, a PhD candidate in the energy and technology policy group at ETH Zurich in Switzerland and one of the authors of the study. (Some experts said it was literally off the scale the researchers gave them.)

The final figure the researchers suggest for fusion’s experience rate is between 2% and 8%, meaning it will see a faster price reduction than nuclear power but not as dramatic an improvement as many common energy technologies being deployed today.

That means that it would take a lot of deployment—and likely quite a long time—for the price of building a fusion reactor to drop significantly, so electricity produced by fusion plants could be expensive for a while. And it’s a much slower rate than the 8% to 20% that many modeling studies assume today.