Forecasting Costs Electrolysers

Abstract

Green hydrogen is increasingly recognised as an essential component for achieving global climate neutrality, offering a versatile and sustainable alternative to fossil fuels in hard-to-abate sectors such as industry, transport, and energy storage. Among the production pathways, water electrolysis using renewable electricity stands out as the most scalable and environmentally viable method. However, a major barrier to large-scale adoption is the uncertainty surrounding future electrolyser costs and deployment trajectories. Most existing cost projections rely on deterministic approaches that provide single-point estimates without explicitly addressing uncertainty, often leading to overly conservative cost decline assumptions and potentially misguided policy and investment decisions.

To address these shortcomings, this thesis develops a probabilistic, empirically grounded framework to forecast both the deployment and capital expenditure (CAPEX) of alkaline electrolysis cells (AEC) and proton exchange membrane (PEM) electrolysers. The framework integrates a logistic S-curve model to simulate technology deployment and a stochastic implementation of Wright’s Law to model cost reductions as a function of cumulative capacity, explicitly incorporating uncertainty through Monte Carlo simulations. Hindcasting validation is used to assess the robustness and predictive performance of the framework, ensuring alignment with historical trends.

The results reveal that achieving the IEA Net Zero Emissions (NZE) 2050 targets would require average annual growth rates of 46% for AEC and 49% for PEM technologies, significantly higher than historical growth trends of around 39%. Even under a more conservative industrial-use scenario, focusing only on sectors such as refining, ammonia, and methanol production, substantial acceleration remains necessary, with required growth rates of 41% for AEC and 45% for PEM. These findings highlight the immense scale of the deployment challenge and the value of explicitly addressing uncertainty when evaluating policy pathways.

Cost forecasts demonstrate a clear divergence between the two technologies. AEC shows a positive experience exponent, suggesting a negative learning rate, implying potential cost increases with expanded deployment. This trend is not statistically significant (p = 0.41), and the model exhibits low explanatory power (R2 = 0.04), indicating that historical data do not support a strong cost-deployment relationship for AEC. In contrast, PEM electrolysers display a statistically significant learning rate of 3.3% (p < 0.01), with an experience exponent of –0.0480 and a higher model fit (R2 = 0.62). Under the reference scenario, AEC median CAPEX is projected to rise to 1,921 EUR/kW by 2030 and 2,076 EUR/kW by 2050, with a wide interquartile range reflecting large uncertainties. For PEM, median CAPEX declines to 1,800 EUR/kW by 2030 and further to 1,533 EUR/kW by 2050, with more pronounced cost reductions under high deployment scenarios.

This transparent and modular framework not only improves cost forecasting for electrolysers but is also adaptable to other emerging energy technologies facing similar learning and deployment uncertainties. By relying on openly available data and explicitly quantifying uncertainty, it provides a robust foundation for analysts, policymakers, and scenario developers. Moreover, its structure makes it suitable for integration into Integrated Assessment Models (IAMs), supporting more realistic and adaptive long-term energy transition planning. Future research should focus on expanding its application across technologies, refining empirical learning rates, and assessing policy impacts within comprehensive system-level analyses, ultimately enabling more confident and informed decisions towards a net-zero future.
The data and the code are available at: https://doi.org/10.4121/82988dc7-099b-45e2-81e2-3850cee1b940