pith. sign in

arxiv: 2604.17421 · v1 · submitted 2026-04-19 · 📡 eess.SY · cs.SY

The structure of technological learning: insights from water electrolysis for cost forecasting, policy, and strategy

Mohamed Atouife , Jesse Jenkins This is my paper

Pith reviewed 2026-05-10 05:54 UTC · model grok-4.3

classification 📡 eess.SY cs.SY
keywords water electrolysislearning curvescost forecastingtechnological learningstructural uncertaintypolicy designindustrial strategysupply chain fragmentation
0
0 comments X

The pith

Plausible modeling choices in technological learning generate widely different cost trajectories for water electrolysis.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper investigates how the structure of technological learning—whether cost reductions are treated as shared across all electrolyzer variants and regions or as fragmented due to competition and supply chain barriers—shapes future cost forecasts. This structural uncertainty matters because learning curves are a primary tool for projecting costs of emerging clean technologies that inform policy, investment, and decarbonization planning. Using water electrolysis as the case, the authors compare shared versus fragmented learning frameworks and show that realistic differences in how knowledge and experience spread produce materially different cost paths. They conclude that relying on any single learning structure risks brittle conclusions for scale-up decisions.

Core claim

Using water electrolysis as a case study, we evaluate how different learning structures, from shared to fragmented learning across technology variants and regions, alter expected cost paths. We interrogate model assumptions that represent contrasting industrial realities, including competition among electrolyzer variants and supply chain fragmentation associated with protectionism and industrial policy. We find that plausible modeling choices generate widely different trajectories, with materially different implications for policy design and technology strategy. We argue for routinely applying multiple learning frameworks to explore decision spaces and stress-test conclusions for scale-up.

What carries the argument

Learning structures ranging from fully shared to fragmented across technology variants and regions, which encode different assumptions about competition, supply-chain integration, and the effects of industrial policy.

If this is right

  • Single learning-curve projections can produce misleading cost estimates for water electrolysis depending on the assumed degree of sharing.
  • Policies that encourage regional or variant-specific supply chains may slow aggregate cost declines relative to fully shared learning scenarios.
  • Technology and investment strategies must incorporate uncertainty over learning structure when planning electrolyzer deployment.
  • Energy-system models should routinely test conclusions against both shared and fragmented learning assumptions for robustness.
  • Decision makers can better explore policy and strategy options by applying multiple learning frameworks rather than a single default.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If real-world learning proves more fragmented than shared, global coordination on deployment could yield larger cost reductions than most current models anticipate.
  • The same structural-uncertainty approach could be applied to other modular clean technologies such as batteries or fuel cells to check whether fragmentation assumptions alter their forecasts.
  • Industrial policies aimed at domestic manufacturing may need explicit trade-offs against the slower cost progress that fragmentation implies.

Load-bearing premise

The contrasted learning structures from shared to fragmented across variants and regions adequately represent real industrial realities including competition and supply-chain fragmentation.

What would settle it

Observing whether future cost reductions in water electrolysis occur uniformly across variants and regions or remain localized and variant-specific over the coming decade would distinguish the shared versus fragmented models.

Figures

Figures reproduced from arXiv: 2604.17421 by Jesse Jenkins, Mohamed Atouife.

Figure 1
Figure 1. Figure 1: The top panel illustrates the evolution of electrolysis stack costs under three learning scenarios. The solid line represents the base case (20% learning rate), while the dashed lines reflect lower (15%) and higher (25%) learning rates. Initial costs are based on a recent Bloomberg New Energy Finance survey36. Shaded regions capture the range of projected costs based on a global installed electrolysis capa… view at source ↗
Figure 2
Figure 2. Figure 2: Learning investment (above) and learning capacity (below) required to reach various Western PEM electrolysis stack cost targets. The continuous line corresponds to the base (20%) learning rate, whereas the dashed lines correspond to low (15%) and high (25%) learning rates. This pattern is especially visible in the case of reducing Western PEM stack costs to 100 USD/kW, a level comparable to the current cos… view at source ↗
Figure 4
Figure 4. Figure 4: BoP and EPC learning curves across different learning models and regions, and projected 2030 costs using BNEF projections. The shaded area and uncertainty bars indicate uncertainty over deployed capacities (varied by 50% relative to the base case) and low (5%) and high (15%) learning rates. Initial costs are based on a recent Bloomberg New Energy Finance survey36 . The implications are also large when expr… view at source ↗
Figure 5
Figure 5. Figure 5: Projected contribution of BoP and EPC costs to LCOH in 2030 across different learning models. The shaded area indicates uncertainty over deployed capacities (varied by 50% relative to the base case) and low (5%) and high (15%) learning rates. Discussion Structural uncertainty is a first-order modeling issue Forecasting the cost evolution of emerging technologies is deeply uncertain not only because learnin… view at source ↗
read the original abstract

Forecasting the cost evolution of emerging clean technologies is crucial for informed policy, investment, and decarbonization decisions, yet it remains deeply uncertain. Learning curves, which link cost declines to cumulative deployment, are widely used for technological cost forecasting. However, applying them to emerging technologies is challenging due to parametric uncertainty in learning rates, which are scarce and highly uncertain, and structural uncertainty stemming from multiple plausible learning frameworks. Using water electrolysis as a case study, we evaluate how different learning structures, from shared to fragmented learning across technology variants and regions, alter expected cost paths. We interrogate model assumptions that represent contrasting industrial realities, including competition among electrolyzer variants and supply chain fragmentation associated with protectionism and industrial policy. We find that plausible modeling choices generate widely different trajectories, with materially different implications for policy design and technology strategy. We argue for routinely applying multiple learning frameworks to explore decision spaces and stress-test conclusions for scale-up planning, national industrial strategy, and energy-systems modeling.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The paper uses water electrolysis as a case study to show that contrasting learning structures—shared versus fragmented across technology variants and regions—generate substantially different cost trajectories. It argues that these structural choices, motivated by industrial realities such as variant competition and supply-chain fragmentation from protectionism, have materially different implications for policy design, technology strategy, and energy-systems modeling, and recommends routinely applying multiple learning frameworks to explore decision spaces and stress-test conclusions.

Significance. If the central result holds, the work usefully highlights structural uncertainty in learning-curve applications for emerging clean technologies and supplies a concrete framework for testing robustness of cost forecasts. This could strengthen decarbonization planning and national industrial strategy by moving beyond single-structure extrapolations.

major comments (2)
  1. [§3.2] §3.2 (Learning Structures): The fragmentation parameters representing supply-chain splits and protectionism are described as interrogating 'contrasting industrial realities,' yet the text provides no calibration or validation against observable data such as regional trade flows, patent citations, or supply-chain maps. Without this grounding, the claim that the modeled structures are plausible rather than illustrative weakens the assertion that the divergent trajectories demonstrate real structural uncertainty dominating policy conclusions.
  2. [§4.3] §4.3 (Cost Trajectory Results): The reported 2030–2050 cost ranges under shared versus fragmented scenarios show clear divergence, but the manuscript does not include a quantitative sensitivity test on the fragmentation intensity parameter itself (e.g., varying it continuously from 0 to 1). This leaves open whether the 'materially different implications' are robust or driven by the specific discrete choices presented.
minor comments (3)
  1. [Abstract] Abstract: The phrase 'scarce and highly uncertain' learning rates is repeated but never quantified with the specific ranges or distributions used in the model runs; adding a short parenthetical or reference to the supplementary material would improve clarity.
  2. [Figure 4] Figure 4: The color scheme distinguishing shared and fragmented trajectories is difficult to distinguish in grayscale; adding line styles or markers would aid readability.
  3. [§2] §2 (Literature Review): The discussion of prior learning-curve applications to electrolysis cites several studies but omits recent work on multi-variant learning (e.g., papers on alkaline vs. PEM competition); a brief addition would strengthen context.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and insightful comments, which help sharpen the presentation of structural uncertainty in learning-curve applications. We address each major comment below and outline the corresponding revisions.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (Learning Structures): The fragmentation parameters representing supply-chain splits and protectionism are described as interrogating 'contrasting industrial realities,' yet the text provides no calibration or validation against observable data such as regional trade flows, patent citations, or supply-chain maps. Without this grounding, the claim that the modeled structures are plausible rather than illustrative weakens the assertion that the divergent trajectories demonstrate real structural uncertainty dominating policy conclusions.

    Authors: We agree that the manuscript would be strengthened by a more explicit discussion of the empirical motivations behind the fragmentation parameters. The parameters are not formally calibrated against trade-flow or patent data because granular, publicly available supply-chain maps for water electrolysis remain limited. Instead, they are constructed to represent contrasting industrial realities documented in recent policy and industry reports on supply-chain localization and protectionism. In the revision we will expand §3.2 to cite these sources, clarify the illustrative yet policy-relevant nature of the scenarios, and note the data constraints that preclude full empirical calibration at present. revision: partial

  2. Referee: [§4.3] §4.3 (Cost Trajectory Results): The reported 2030–2050 cost ranges under shared versus fragmented scenarios show clear divergence, but the manuscript does not include a quantitative sensitivity test on the fragmentation intensity parameter itself (e.g., varying it continuously from 0 to 1). This leaves open whether the 'materially different implications' are robust or driven by the specific discrete choices presented.

    Authors: We accept that a continuous sensitivity analysis on the fragmentation intensity parameter would improve the robustness demonstration. We will add this analysis to §4.3, sweeping the parameter from 0 (fully shared learning) to 1 (fully fragmented) and showing the resulting cost trajectories. This will confirm that the qualitative divergence in policy implications persists across intermediate fragmentation levels rather than arising solely from the discrete endpoints. revision: yes

Circularity Check

0 steps flagged

No circularity: structural sensitivity analysis remains independent of fitted inputs.

full rationale

The paper presents a sensitivity study comparing shared versus fragmented learning structures across electrolyzer variants and regions, using water electrolysis as a case study to show that different plausible structures produce divergent cost trajectories. No equations, parameter-fitting procedures, or self-citations are described in the provided text that would reduce any claimed prediction or result to the inputs by construction. The central exercise is framed as interrogating modeling assumptions that represent contrasting industrial realities, without evidence of self-definitional loops, fitted parameters relabeled as predictions, or load-bearing self-citations. The derivation chain is therefore self-contained as an exploratory comparison rather than a closed tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities; the central claim rests on the unstated assumption that alternative learning structures can be parameterized from existing data.

pith-pipeline@v0.9.0 · 5468 in / 1038 out tokens · 30608 ms · 2026-05-10T05:54:02.065933+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages

  1. [1]

    Wright, T. P. Factors Affecting the Cost of Airplanes. J. Aeronaut. Sci. 3, 122–128 (1936)

    work page 1936

  2. [2]

    Colpier, U. C. & Cornland, D. The economics of the combined cycle gas turbine—an experience curve analysis. Energy Policy 30, 309–316 (2002)

    work page 2002

  3. [3]

    & Ménière, Y

    de La Tour, A., Glachant, M. & Ménière, Y. Predicting the costs of photovoltaic solar modules in 2020 using experience curve models. Energy 62, 341–348 (2013)

    work page 2020

  4. [4]

    & Turkenburg, W

    Junginger, M., Faaij, A. & Turkenburg, W. C. Global experience curves for wind farms. Energy Policy 33, 133–150 (2005)

    work page 2005

  5. [5]

    S., Yeh, S., Antes, M., Berkenpas, M

    Rubin, E. S., Yeh, S., Antes, M., Berkenpas, M. & Davison, J. Use of experience curves to estimate the future cost of power plants with CO2 capture. Int. J. Greenh. Gas Control 1, 188–197 (2007)

    work page 2007

  6. [6]

    Gan, P. Y. & Li, Z. Quantitative study on long term global solar photovoltaic market. Renew. Sustain. Energy Rev. 46, 88–99 (2015)

    work page 2015

  7. [7]

    & Söderholm, P

    Lindman, Å. & Söderholm, P. Wind power learning rates: A conceptual review and meta- analysis. Energy Econ. 34, 754–761 (2012)

    work page 2012

  8. [8]

    Photovoltaic learning rate estimation: Issues and implications

    Mauleón, I. Photovoltaic learning rate estimation: Issues and implications. Renew. Sustain. Energy Rev. 65, 507–524 (2016)

    work page 2016

  9. [9]

    Use of experience curves to analyse the prospects for diffusion and adoption of renewable energy technology

    Neij, L. Use of experience curves to analyse the prospects for diffusion and adoption of renewable energy technology. Energy Policy 25, 1099–1107 (1997)

    work page 1997

  10. [10]

    The experience curve theory and its application in the field of electricity generation technologies – A literature review

    Samadi, S. The experience curve theory and its application in the field of electricity generation technologies – A literature review. Renew. Sustain. Energy Rev. 82, 2346–2364 (2018)

    work page 2018

  11. [11]

    & Staffell, I

    Schmidt, O., Hawkes, A., Gambhir, A. & Staffell, I. The future cost of electrical energy storage based on experience rates. Nat. Energy 2, 1–8 (2017)

    work page 2017

  12. [12]

    Experience Curves for Energy Technology Policy

    International Energy Agency. Experience Curves for Energy Technology Policy. (OECD, 2000). doi:10.1787/9789264182165-en

    work page doi:10.1787/9789264182165-en 2000

  13. [13]

    Mayer, J. et al. Current and Future Cost of Photovoltaics. (2015) doi:10.13140/RG.2.1.1371.7206

    work page doi:10.13140/rg.2.1.1371.7206 2015

  14. [14]

    & Auer, H

    Haas, R., Sayer, M., Ajanovic, A. & Auer, H. Technological learning: Lessons learned on energy technologies. WIREs Energy Environ. 12, e463 (2023)

    work page 2023

  15. [15]

    R., Yip, A

    Lovering, J. R., Yip, A. & Nordhaus, T. Historical construction costs of global nuclear power reactors. Energy Policy 91, 371–382 (2016)

    work page 2016

  16. [16]

    S., Ossenbrink, J

    Huenteler, J., Schmidt, T. S., Ossenbrink, J. & Hoffmann, V. H. Technology life-cycles in the energy sector — Technological characteristics and the role of deployment for innovation. Technol. Forecast. Soc. Change 104, 102–121 (2016)

    work page 2016

  17. [17]

    How Solar Energy Became Cheap: A Model for Low-Carbon Innovation

    Nemet, G. How Solar Energy Became Cheap: A Model for Low-Carbon Innovation

    work page

  18. [18]

    & Trancik, J

    Kavlak, G., McNerney, J. & Trancik, J. E. Evaluating the causes of cost reduction in photovoltaic modules. Energy Policy 123, 700–710 (2018)

    work page 2018

  19. [19]

    M., Kavlak, G., McNerney, J

    Klemun, M. M., Kavlak, G., McNerney, J. & Trancik, J. E. Mechanisms of hardware and soft technology evolution and the implications for solar energy cost trends. Nat. Energy 8, 827–838 (2023)

    work page 2023

  20. [20]

    & Schmidt, T

    Huenteler, J., Niebuhr, C. & Schmidt, T. S. The effect of local and global learning on the cost of renewable energy in developing countries. J. Clean. Prod. 128, 6–21 (2016)

    work page 2016

  21. [21]

    & Rubin, E

    Yeh, S. & Rubin, E. S. A review of uncertainties in technology experience curves. Energy Econ. 34, 762–771 (2012)

    work page 2012

  22. [22]

    IRENA (2020), Green Hydrogen Cost Reduction: Scaling up Electrolysers to Meet the 1.50C Climate Goal, International Renewable Energy Agency, Abu Dhabi

    work page 2020

  23. [23]

    & Vanderzwaan, B

    Schoots, K., Ferioli, F., Kramer, G. & Vanderzwaan, B. Learning curves for hydrogen production technology: An assessment of observed cost reductions. Int. J. Hydrog. Energy 33, 2630–2645 (2008)

    work page 2008

  24. [24]

    & Möst, D

    Bühler, L. & Möst, D. Derivation of one- and two-factor experience curves for electrolysis technologies. Int. J. Hydrog. Energy 89, 105–116 (2024)

    work page 2024

  25. [25]

    & Reichelstein, S

    Glenk, G., Meier, R. & Reichelstein, S. Cost Dynamics of Clean Energy Technologies. Schmalenbach J. Bus. Res. 73, 179–206 (2021)

    work page 2021

  26. [26]

    & Gomonov, K

    Revinova, S., Lazanyuk, I., Ratner, S. & Gomonov, K. Forecasting Development of Green Hydrogen Production Technologies Using Component-Based Learning Curves. Energies 16, (2023)

    work page 2023

  27. [27]

    H., Thomassen, M

    Reksten, A. H., Thomassen, M. S., Møller-Holst, S. & Sundseth, K. Projecting the future cost of PEM and alkaline water electrolysers; a CAPEX model including electrolyser plant size and technology development. Int. J. Hydrog. Energy 47, 38106–38113 (2022)

    work page 2022

  28. [28]

    & Reichelstein, S

    Glenk, G., Holler, P. & Reichelstein, S. Advances in power-to-gas technologies: cost and conversion efficiency. Energy Environ. Sci. 16, 6058–6070 (2023)

    work page 2023

  29. [29]

    & Zauner, A

    Böhm, H., Goers, S. & Zauner, A. Estimating future costs of power-to-gas – a component-based approach for technological learning. Int. J. Hydrog. Energy 44, 30789– 30805 (2019)

    work page 2019

  30. [30]

    & Jenkins, J

    Atouife, M. & Jenkins, J. D. Learning-Driven Economics of Electrolytic Hydrogen Dashboard. https://learning-driven-economics-of-electrolytic-hydrogen- dashboard.streamlit.app/

    work page

  31. [31]

    & van der Zwaan, B

    Ferioli, F., Schoots, K. & van der Zwaan, B. C. C. Use and limitations of learning curves for energy technology policy: A component-learning hypothesis. Energy Policy 37, 2525– 2535 (2009)

    work page 2009

  32. [32]

    J., Junginger, M

    Krishnan, S., Corona, B., Kramer, G. J., Junginger, M. & Koning, V. Prospective LCA of alkaline and PEM electrolyser systems. Int. J. Hydrog. Energy 55, 26–41 (2024)

    work page 2024

  33. [33]

    https://www.siemens- energy.com/us/en/home/stories/electrolyzer-gigawatt-factory.html

    The race to ramp up renewable green hydrogen goes high tech. https://www.siemens- energy.com/us/en/home/stories/electrolyzer-gigawatt-factory.html

    work page

  34. [34]

    virtuous cycle

    Watanabe, C., Wakabayashi, K. & Miyazawa, T. Industrial dynamism and the creation of a “virtuous cycle” between R&D, market growth and price reduction: The case of photovoltaic power generation (PV) development in Japan. Technovation 20, 299–312 (2000)

    work page 2000

  35. [35]

    P., He, G

    Helveston, J. P., He, G. & Davidson, M. R. Quantifying the cost savings of global solar photovoltaic supply chains. Nature 612, 83–87 (2022)

    work page 2022

  36. [36]

    Electrolyzer Price Survey 2024 Rising Costs, Glitchy Tech

    BloombergNEF. Electrolyzer Price Survey 2024 Rising Costs, Glitchy Tech. (2024)

    work page 2024

  37. [37]

    Electric Hydrogen https://eh2.com/

    High-power electrolyzers for the lowest cost clean hydrogen. Electric Hydrogen https://eh2.com/

    work page

  38. [38]

    FROM AMBITION TO REALITY 3

    Worley. FROM AMBITION TO REALITY 3. https://www.worley.com/en/insights/our- thinking/energy-transition/from-ambition-to-reality/ (2022)

    work page 2022

  39. [39]

    & Schmidt, T

    Malhotra, A. & Schmidt, T. S. Accelerating Low-Carbon Innovation. Joule 4, 2259–2267 (2020)

    work page 2020

  40. [40]

    White, J. B. (CONTR). Comparison of Commercial, State-of-the-Art, Fossil-Based Hydrogen Production Technologies

    work page

  41. [41]

    Penev, M. et al. Capital Structure for Techno-Economic Analysis of Hydrogen Projects. NREL/TP--5400-90103, 2397248, MainId:91881 https://www.osti.gov/servlets/purl/2397248/ (2024) doi:10.2172/2397248

    work page doi:10.2172/2397248 2024

  42. [42]

    M., Albardi, S

    Elshurafa, A. M., Albardi, S. R., Bigerna, S. & Bollino, C. A. Estimating the learning curve of solar PV balance–of–system for over 20 countries: Implications and policy recommendations. J. Clean. Prod. 196, 122–134 (2018)

    work page 2018

  43. [43]

    A Hydrogen Strategy for a Climate-Neutral Europe

    European Comission. A Hydrogen Strategy for a Climate-Neutral Europe. https://energy.ec.europa.eu/system/files/2020-07/hydrogen_strategy_0.pdf/ (2020)

    work page 2020

  44. [44]

    & Steinfeld, E

    Nahm, J. & Steinfeld, E. S. Scale-up Nation: China’s Specialization in Innovative Manufacturing. World Dev. 54, 288–300 (2014)

    work page 2014

  45. [45]

    & Nahm, J

    Helveston, J. & Nahm, J. China’s key role in scaling low-carbon energy technologies. Science 366, 794–796 (2019)

    work page 2019

  46. [46]

    China’s Specialization in Innovative Manufacturing

    Nahm, J. China’s Specialization in Innovative Manufacturing. in Collaborative Advantage: Forging Green Industries in the New Global Economy (ed. Nahm, J.) 0 (Oxford University Press, 2021). doi:10.1093/oso/9780197555361.003.0005

    work page doi:10.1093/oso/9780197555361.003.0005 2021

  47. [47]

    Cost of Capital Observatory

    IEA. Cost of Capital Observatory. https://www.iea.org/reports/cost-of-capital- observatory (2023)

    work page 2023