Hydrogen cost forecasts often rely on the experience curve, a key tool in clean-energy transitions. This method estimates future costs by tracking how much a technology’s price drops each time cumulative deployment doubles. It works best when each doubling of installed capacity reflects a similar increase in manufacturing and installation of comparable units. However, as discussed in my earlier analysis, electrolyzers and hydrogen plants differ significantly from solar modules and batteries in this regard. Recent evidence highlights another issue: installed gigawatts can increase by making stacks, modules, and plants larger, so a doubling in capacity does not always equate to a doubling in manufacturing experience.
This distinction is clear in a 2025 European study of electrolyzer projects, which compiled capital-cost and capacity data from 2005 onward. The raw data show impressive experience rates: costs fall 23.3% across all projects, 32.1% for PEM, and 22.9% for alkaline electrolysis with each doubling of installed capacity. These rates appear similar to or better than those for solar and batteries. However, when project costs are adjusted for estimated economies of scale, the rates drop to 13.3%, 17.6%, and 7.3% respectively, with the alkaline relationship losing statistical significance. While costs did decline, the raw curve attributed several different mechanisms to “learning.”
The core issue is what a doubling truly represents. The full TFIE Strategy Briefing analysis examines stack-size growth, chemical-plant scale, manufacturing repetition, and the system boundary needed for a reliable hydrogen-cost forecast.
Chemical-plant scale economies are significant but mostly realized early. A 100 MW hydrogen plant does not require one hundred times the compressors, transformers, water-treatment systems, cooling equipment, gas purification, or engineering of a 1 MW plant. Larger projects share equipment, distribute fixed engineering costs, and optimize common systems, leading to dramatic cost reductions as projects scale up. Once equipment reaches practical scale, further capacity increases come from replicating optimized process trains. New plants still benefit from better procurement, standardized designs, and more experienced contractors, but they do not repeatedly capture the full gains from moving beyond demonstration scale.
The denominator becomes more complex when stack sizes increase. For example, if cumulative electrolysis capacity grows from 5 GW to 50 GW, that is a tenfold increase, or 3.32 doublings. If the average stack remains 1 MW, the number of stacks also increases tenfold, so capacity and stack-count doublings align. However, if the average stack size grows from 1 MW to 5 MW, only 10,000 stacks are needed at 50 GW. Installed capacity still shows 3.32 doublings, but the stack count has only doubled once. Factories gain experience, but not at the rate suggested by cumulative gigawatts.
Stack count is not a perfect measure of manufacturing experience either. Larger stacks contain repeated cells, membranes, plates, and electrode areas, each with its own manufacturing learning. Improved materials and higher current density can boost output without proportional increases in material. The key point is that during rapid equipment upscaling, installed megawatts can overstate the actual manufacturing repetitions at several levels. An experience curve based solely on cumulative GW overlooks these changes in physical architecture.
The system boundary adds another constraint. According to the IEA’s 2025 electrolyzer cost breakdown, the stack accounts for only about 15-20% of installed capital cost. Around 25-30% is in balance-of-plant equipment like power electronics, piping, compressors, and gas treatment, while engineering, procurement, construction, and contingency can make up more than half. Even a 20% cost reduction in a component that is one fifth of total CAPEX only reduces the overall project cost by about 4%. Other project elements can improve as well, but compressors, civil works, electrical connections, and construction follow their own scale and productivity paths, separate from the stack factory’s learning rate.
This does not mean electrolyzer projects will remain as costly as today’s first-generation installations. Manufacturers can enhance stack designs and materials, EPC contractors can standardize layouts, procurement can become more efficient, shared equipment can reach optimal scale, and repeated construction can eliminate first-of-a-kind errors. The mistake is to combine all these mechanisms into a single historical learning rate per installed-capacity doubling and project that rate forward to 2035 or 2050. Many of the largest early cost reductions occur because the industry is transitioning from small demonstration projects to fully scaled industrial facilities.
Electrolyzer CAPEX is only one part of the final hydrogen price. Electricity is the main variable cost, and manufacturing improvements cannot eliminate it. Very cheap wind and solar power is often intermittent, while capital equipment benefits from high utilization. Compression, storage, and distribution follow, each with its own infrastructure economics. Better equipment will lower costs, but not every part of the production and delivery chain can be mass-produced like solar modules.
A robust hydrogen forecast must separate manufacturing improvements, electrochemical performance, stack-size effects, chemical-plant scale, repeat engineering and construction, electricity, utilization, financing, and logistics, modeling each on its own terms. Hydrogen will become cheaper than many early projects, but recent evidence suggests there is less reason to expect a steep, persistent cost curve simply because global electrolyzer capacity keeps doubling. Low-carbon hydrogen should be planned around the realistic prices complete systems can achieve and targeted toward uses where the molecule is valuable enough to justify those costs.
For a deeper analysis of what apparent learning rates are actually measuring - including stack-size growth, project-scale effects, manufacturing repetition, and the system boundary needed for a reliable hydrogen-cost forecast - refer to the full analysis in TFIE Strategy Briefing.
.svg)
