Fusion Power Needs 34 Capacity Doublings to Halve Its Cost. Solar Needed Three.
A peer-reviewed Nature Energy study from ETH Zurich estimates fusion power's experience rate at 2 to 8 percent, far below the 8 to 20 percent that investors, energy modelers, and startup pitch decks have assumed for decades. At the low end of that range, fusion would need 34 doublings of cumulative deployed capacity before its cost drops by half. Solar PV needed three. Fifteen billion dollars in private fusion capital is priced to a learning curve that the underlying technology physics may not support.
Thirty-four. That is the number of times cumulative fusion capacity must double before the technology's cost drops by half, if the low end of a new ETH Zurich estimate proves correct. Solar photovoltaics, the technology that fell 90 percent in cost over the past 15 years and upended every long-range energy forecast published by the International Energy Agency during that period, required just three doublings to achieve the same halving. Three. The difference is not incremental but structural, and a team of researchers from the Swiss Federal Institute of Technology has now published the peer-reviewed math to prove it.
Tang, Noll, and Schmidt (2026), writing in Nature Energy, linked three technological characteristics to empirical experience rates using decades of innovation literature and supplementary expert interviews. Their conclusion: fusion power plants have an experience rate of 2 to 8 percent, depending on reactor concept, with magnetic confinement and laser inertial confinement clustering at the low end, which is precisely where the commercial money sits. Current energy system models and investor projections routinely assume 8 to 20 percent. Those numbers are borrowed from other technologies or conjured from thin air. One expert the team interviewed described fusion's design complexity as "literally off the scale."
The Math Nobody Ran
Experience rates govern cost learning: an experience rate of 23 percent means that every time cumulative deployed capacity doubles, cost per unit drops by 23 percent, so the formula for how many doublings it takes to halve cost is straightforward: n = ln(0.5) / ln(1 - ER). Run the formula across every major energy technology, and the divergence is devastating.
| Technology | Experience Rate | Doublings to Halve Cost |
|---|---|---|
| Solar PV | 23% | 2.6 |
| Li-ion batteries | 20% | 3.1 |
| Onshore wind | 12% | 5.4 |
| Fusion (optimistic) | 8% | 8.3 |
| Nuclear fission | 2% | 34.3 |
| Fusion (pessimistic) | 2% | 34.3 |
At 23 percent, solar needed roughly 2.6 doublings, and lithium-ion batteries at 20 percent needed about 3.1, which is why EV battery packs that cost $1,200 per kilowatt-hour in 2010 fell below $120 in barely a decade. These are technologies where you build a thing in a factory, stamp out copies, ship them in containers, and learn from every iteration at manufacturing speed. Fusion's optimistic scenario of 8 percent would still require 8.3 doublings, and its pessimistic scenario lands at 34.3, identical to nuclear fission, a technology whose costs in several countries have actually increased with scale.
What 10 Doublings Cost the World
Doublings sound abstract until you translate them into real capacity. Then the numbers become alarming. If the first commercial fusion plant is 500 MW, then 10 doublings equal 512 GW of deployed fusion, approximately 6 percent of global electricity capacity, a buildout that would take decades even under the most aggressive deployment scenarios ever attempted for any energy technology. At 2 percent experience rate, those 10 doublings would reduce cost by just 18.3 percent. Not half. Eighteen percent. At solar's 23 percent rate applied over the same 10 doublings, cost would fall by 92.7 percent, which is why solar went from boutique curiosity to the cheapest electricity source in human history within a single generation while nuclear fission, which started commercializing in the 1950s, still struggles to compete on cost seven decades later.
Most fusion startups target a levelized cost of electricity between $50 and $80 per megawatt-hour for their first commercial plants, which sounds competitive until you examine the capital expenditure required to build the reactors that produce those megawatt-hours. First-of-a-kind estimates range from $8,000 to $15,000 per kilowatt, while solar installations now cost $500 to $800 per kilowatt, a gap so wide that no plausible learning curve closes it within the investment horizon of any fund currently backing fusion. Even at the 8 percent experience rate, closing that gap requires deploying fusion at a scale nobody has seriously modeled, because the industry has not yet produced a single net-energy-positive reactor. Not one. Let alone a commercial power plant.
Where Private Capital Went
Fifteen billion dollars in private money has flowed into fusion startups, according to WebProNews and industry trackers. Commonwealth Fusion Systems alone has raised $3 billion, roughly a third of all private fusion funding in history. Helion Energy closed a $425 million Series F on a $5.4 billion valuation and signed a power purchase agreement with Microsoft. TAE Technologies has raised over $1.2 billion across multiple rounds. The money keeps coming. The US Department of Energy allocated $1.01 billion to fusion in fiscal year 2024, nearly five times its $216 million geothermal budget, despite the fact that geothermal produces commercial electricity today and fusion does not.
ITER tells the story in miniature, the purest demonstration of what a 2 percent experience rate looks like when bolted to a $22 billion construction schedule. The international tokamak being built in Cadarache, France, was budgeted at $5 billion in 2006 and now exceeds $22 billion, a cost overrun of roughly 340 percent that has become so routine in fusion discourse that it barely registers as news anymore. First plasma has been pushed from 2018 to 2025 to 2034, a 16-year schedule slip on a project that the international physics community once described as a decade away from first results. Remember: ITER is a research device, one that will never deliver a single commercial watt to any grid on Earth. Its central solenoid magnet was delivered in May 2026 to a facility that will not produce fusion conditions for another eight years. That cost trajectory is not an anomaly. It is exactly what a 2 percent experience rate predicts for large, bespoke, extraordinarily complex construction projects.
The Strongest Case for Fusion Anyway
Egemen Kolemen, a plasma physicist at Princeton's Plasma Physics Laboratory, told MIT Technology Review the ETH Zurich study might be making the same mistake solar pessimists made in 2000. "Many analysts predicted that solar power would remain expensive, but then production exploded and prices came crashing down, largely because China went all in," Kolemen said. "We haven't built the thing yet, so we don't know."
Fair point. This is the strongest counterargument, and it deserves serious engagement rather than dismissal, because solar's trajectory genuinely surprised nearly everyone, including the forecasters at the International Energy Agency who underestimated solar deployment by orders of magnitude for over a decade. If a government with sufficient industrial capacity decided to build fusion reactors at national scale regardless of near-term economics, the resulting production experience could drive cost reductions that no model grounded in current data would predict.
But the ETH Zurich team specifically addressed this objection, and their rebuttal rests on the three technological characteristics that most reliably predict experience rates in the empirical literature. Solar's breakout was driven by one word: modularity. A solar panel is a kilowatt-scale object manufactured in a semiconductor fab, shipped in a box, and installed by two workers with hand tools. Every unit benefits from process improvements propagated across millions of identical copies. Fusion cannot. A tokamak is a gigawatt-scale bespoke construction project involving superconducting magnets, neutron-absorbing blankets, plasma-facing materials operating at 150 million degrees Celsius, and first-wall components that degrade under neutron bombardment and must be replaced on timelines nobody has measured in a production environment. The unit size, design complexity, and customization requirements that predict low experience rates in the innovation literature describe fusion with uncomfortable precision. No amount of political will changes the physics of building a reactor at scale, because the constraint is not money or ambition but the irreducible complexity of confining plasma at 150 million degrees Celsius inside a machine that can withstand neutron bombardment for decades without structural failure. China can subsidize the cost. It cannot subsidize away the complexity.
What This Analysis Did Not Prove
The Tang et al. study estimates experience rates by analogy with other large, complex energy technologies, supplemented by expert interviews. It does not derive rates from actual fusion deployment data, because no such data exists anywhere on Earth. None. Nobody has ever built a commercial fusion power plant, which means the experience rate is a prediction informed by analogy and expert judgment, not a measurement derived from observed cost curves. If a technological discontinuity occurs, like the transition from tokamaks to a simpler reactor architecture, the rate could shift, because stellarators, laser fusion, and compact high-field magnets represent different design philosophies with different complexity profiles, and lumping them into a single 2 to 8 percent band may understate the variance.
The $15 billion private investment figure aggregates fundraising across startups at wildly different stages, and some of that capital funds fundamental plasma physics research that may never translate into deployable cost-reduction data because it never touched a production line where experience curves actually compound. Comparing it directly to solar investment, which funded manufacturing scale, conflates R&D spending with production learning.
Additionally, the US DOE recently signed a 10-year partnership with the Max Planck Institute for the Wendelstein 7-X stellarator, which achieved a 43-second sustained fusion pulse in 2025 at a total site cost of 1.44 billion euros. Stellarators may have fundamentally different scaling properties than tokamaks, and the field is too immature for anyone to know.
What You Can Do
If you allocate capital to clean energy, whether as a fund manager, a utility procurement officer, or a policymaker writing R&D budgets, demand that fusion cost projections specify their assumed experience rate and justify it against the Tang et al. range of 2 to 8 percent. Any projection using 15 or 20 percent should explain why fusion will behave more like solar panels than like nuclear fission reactors when the underlying technology characteristics, specifically unit size, design complexity, and customization requirements, map far more closely to fission.
If you hold equity in a fusion startup, recalculate your internal rate of return using the 2 percent floor and compare it to the 8 to 20 percent your original investment thesis assumed, because the difference between halving cost in 8 doublings versus 34 doublings is not a rounding error. It determines whether fusion reaches grid parity in 2045 or 2085, a gap of an entire generation.
If you are a fusion researcher, this study does not argue against building fusion; it argues against assuming fusion will become cheap on the same timeline as modular technologies. The honest response is to pursue reactor designs that maximize modularity and minimize bespoke construction, because those are the characteristics that predict high experience rates. Commonwealth Fusion Systems' compact high-field magnet approach and Helion's pulsed-field design represent attempts to move fusion toward smaller, more manufacturable units, and whether they succeed determines whether fusion's experience rate lands at 2 or 8. That fourfold difference is the entire ballgame for the future of clean energy.
The Bottom Line
A Nature Energy study from three ETH Zurich researchers quantifies what fusion skeptics have argued for decades: the technology's cost-reduction trajectory looks nothing like solar's. Fusion's experience rate of 2 to 8 percent means the industry needs between 8 and 34 capacity doublings to halve its cost. Solar needed three. Lithium-ion batteries needed about three. These are not technologies waiting for a breakthrough; they are technologies that already broke through because their physical characteristics, modular, mass-manufacturable, and factory-learned, enabled rapid cost learning. Fusion's characteristics, enormous unit size, extraordinary complexity, and site-specific construction, predict the opposite. Fifteen billion dollars in private capital and over a billion annually in public funding are betting that fusion's learning curve will defy the pattern observed across every other large-scale bespoke energy technology in history. The ETH Zurich team has given that bet a number. Just one. It is 2 to 8 percent, and everything else follows from the math.