The green hydrogen ambition and implementation gap

Overview

Our approach was split into three parts. First, we tracked green hydrogen project announcements to quantify the green hydrogen implementation gap in 2022 and 2023. Second, we compared project announcements with 1.5 °C scenarios to show the 2030 green hydrogen ambition gap. Third, we modelled the pay-as-bid market premium and estimated required subsidies using a competitiveness analysis of four green products and five fossil competitors across 14 end uses, which led to the 2030 green hydrogen implementation gap.

Green hydrogen projects database

We used data of electrolysis project announcements from the IEA Hydrogen Production Projects and Infrastructure Database⁵⁸ (previously called the IEA Hydrogen Projects Database), incorporating three database snapshots from 2021, 2022 and 2023. We only included project announcements for electrolysers that included a year of project launch, had a meaningful status (not ‘Other’ or ‘Other/Unknown’) and reported a capacity value. We did not filter for the type of electricity as this was often unknown. These criteria led to 612 projects in the 2021 snapshot, 877 projects in the 2022 snapshot and 1,265 projects in the 2023 snapshot. In the 2023 snapshot, only a single status category was reported for projects that were either under construction or had an FID (‘FID/Construction’). To ensure consistent status categories across all snapshots, we merged the ‘FID’ and ‘Under construction’ categories in the 2021 and 2022 snapshots. Projects with a ‘DEMO’ status were allocated as ‘Operational’, ‘FID/Construction’ or ‘Decommissioned’, depending on whether they were still running, announced for the future or had been decommissioned, respectively. We note that the ‘Concept’ category is very broadly defined with an unspecified credibility bar for inclusion, while the ‘Feasibility study’ category may also contain projects for which a feasibility study is planned, but has not yet started. Confidential projects were distributed to all regions in proportion to the share of capacity from non-confidential projects, but could not be tracked across database snapshots.

Data quality validation

We conducted a comprehensive, structured and fully documented data quality validation of the green hydrogen project announcements, manually validating 524 project entries across all three database versions. For projects announced for 2022 or 2023, we covered at least 90% of the announced capacity, while for projects announced for 2024–2030, we covered at least 75% of the announced capacity in all three database versions (Supplementary Table 1). In addition, we manually verified the fate of all projects announced to launch in 2023 in the database published in October 2023 (Fig. 3). Note that we did not attempt to identify missing projects, implying that the success rate may change if projects that were realized in 2023 were missing from the most recent database version included in this analysis, published in October 2023. During the data validation, we adjusted the size of a project if it was not operating at its nameplate capacity, which was the case for the world’s largest green hydrogen project, Sinopec Kuqa in China. The data quality validation procedure is described in detail in Supplementary Note 1.

Tracking green hydrogen projects

Each project has a unique reference number that stays the same across all database snapshots, as confirmed by the IEA in personal correspondence. This enabled us to track the development of project announcements over time (see Fig. 3 for projects announced for 2023, Supplementary Fig. 5 for projects announced for 2022 and Supplementary Fig. 6 for projects announced for 2024). Supplementary Figs. 7–10 also show the 2023 project tracking for those regions that have at least ten trackable project entries. We accounted for changing capacity of projects between two database snapshots by adding dummy projects, which are, however, not explicitly shown in the Sankey diagrams for simplicity. The reported rates of disappearance, delay and success (Fig. 3b–d and Supplementary Fig. 5b, c) only refer to projects announced in 2021, 2022 and 2023, respectively.

Green hydrogen in 1.5 °C scenarios

As an indicator of green hydrogen requirements in stringent climate mitigation scenarios, we collected electrolysis capacity values from a wide range of 1.5 °C scenarios, including (1) IAM scenarios and (2) institutional and corporate scenarios (Extended Data Fig. 1). For the IAM scenarios, we used the IPCC AR6 Scenarios Database⁵⁹ (category C1) as well as the Network for Greening the Financial System (NGFS) dataset⁶⁰ (Version 4.2, the ‘Net Zero 2050’ and ‘Low demand’ scenarios). We excluded IAM scenarios that always report zero electrolysis capacity (or zero electrolytic hydrogen production) or, in any period from 2025, report a value that is lower than the operational electrolysis capacity in 2023. We also omitted scenarios from the NGFS project that included climate damages as this is only reported by one model. For the institutional and corporate scenarios, due to limited reporting of numerical data in text or tables, in some cases we resorted to extracting data from graphics using WebPlotDigitizer, which has been shown to be reliable⁶¹. All datasets are available via GitHub (see the Data availability statement). If electrolysis capacity was not directly reported, we converted production quantities into electrolysis capacity, assuming 3,750 full load hours, 69% efficiency and the lower heating value of hydrogen, 33.33 kWh kg⁻¹. For IAM scenarios, we transformed the reported hydrogen output capacity to the corresponding input capacity of the electrolyser using the efficiency of 69%. Due to these approximations, reported electrolysis requirements in 1.5 °C scenarios are inherently uncertain.

Modelling pay-as-bid market premiums

To quantify the future green hydrogen implementation gap, we developed a model of the required pay-as-bid market premiums for green hydrogen projects (Extended Data Fig. 3). First, we mapped each of the 14 end-use categories from the green hydrogen projects database to the competition between a green product and a fossil competitor, covering four green products (green hydrogen, e-methanol, e-kerosene and e-methane) and five fossil competitors (grey hydrogen, natural gas, grey methanol, diesel and kerosene), as shown in Extended Data Table 1. For projects without a designated end use, we assumed that green hydrogen competes with natural gas. Second, we calculated the levelized cost of all green products (Extended Data Table 2) and the prices of all fossil competitors with and without an ambitious carbon price pathway that is in line with EU climate targets⁴¹ (Extended Data Table 3). Details on these costs and prices are explained in the following sections. Third, we incorporated demand-side policies such as end-use quotas, which increase the willingness to pay for green products and thereby reduce required policy costs (Supplementary Fig. 15). Finally, for each end use, we estimated the required subsidies based on (1) vintage tracking of project announcements and (2) the cost gap between the green product and the fossil competitor (Extended Data Fig. 3).

We included global estimates of implemented demand-side policies in 2030 across four end uses, provided by the IEA¹, which we converted into the corresponding electrolysis capacities using the lower heating value, as well as the full load hours and efficiencies of the respective scenario. We proportionally distributed these estimates of electrolysis capacity that are supported by demand-side regulation in 2030 according to the project announcements from 2024–2030 (Supplementary Fig. 15).

If the capacity supported by demand-side policies exceeded the announced capacity, which is the case for refining and synthetic fuels, we omitted the difference, assuming that demand-side policies are end-use specific.

To estimate the required annual subsidies, we combined these components. As shown in Fig. 5a–d and Extended Data Fig. 5, for each end use, the instantaneous cost gap (Δp_t) between the levelized cost of the green product in year t (LCOX_t) and the price of the fossil competitor (\({p}_{t}^{{{\rm{fossil}}}}\)) is given as:

However, this cannot be used directly to estimate subsidies. As illustrated in Extended Data Fig. 3, a green hydrogen or electrofuel project completed in year t′ must sell the green product at \({{\rm{LCOX}}}_{t^{\prime}}\) for the entire duration of the payback period τ to recover its costs. The required annual subsidies accumulate over time due to projects built in previous years. For example, in 2026, projects that were built in 2024 face a cost gap of \({{\rm{LCOX}}}_{2024}-{p}_{2026}^{{{\rm{fossil}}}}\), projects that were built in 2025 face a cost gap of \({{{\rm{LCOX}}}}_{2025}-{p}_{2026}^{{{\rm{fossil}}}}\) and projects that were built in 2026 face a cost gap of \({{{\rm{LCOX}}}}_{2026}-{p}_{2026}^{{{\rm{fossil}}}}\). These cost gaps have to be bridged for the electrolysis capacity built in the corresponding year t′, denoted as \(\Delta C_{t ^{\prime}}\) (accounting for capacity supported by demand-side policies). For each end use, with electrolyser full load hours \({{\rm{FLH}}}_{{\rm{H}}_2}\), electrolyser efficiency \({\eta }_{{\rm{H}}_2}\) and payback period τ, the required annual subsidy (\({S}_{t}^{{{\rm{annual}}}}\)) in year t is given as:

Note that for subsidies in year t, only the price of the fossil competitor (\({p}_{t}^{{{\rm{fossil}}}}\)) refers to the same year t, whereas all other parameters refer to the year t′ in which the project was built. Thus, the realization of green hydrogen projects built in the year t′ requires subsidy payments for the full payback period \([t^{\prime},\;t^{\prime} +\tau)\) as long as \({{{\rm{LCOX}}}}_{t^{\prime} } > {p}_{t}^{{{\rm{fossil}}}}\). For end uses where the green product and the fossil competitor are not used thermally, we included the relative efficiency improvement of using the green product over the fossil competitor, \({\eta }_{{{\rm{LHV}}}}^{{{\rm{green}}}}/{\eta }_{{{\rm{LHV}}}}^{{{\rm{fossil}}}}\), adjusting the LCOX accordingly (Extended Data Table 1). Note that for green hydrogen, we denote LCOX as LCOH. Correspondingly, the required cumulative subsidies until year t (\({S}_{t}^{{{\rm{cumulative}}}}\)) are given by:

We show in Fig. 5e–g and Extended Data Fig. 6 the required annual and cumulative subsidies as the sum over all end uses.

To analyse what would be required for a 1.5 °C scenario, after 2030 we used the median of the institutional and corporate 1.5 °C scenarios for \({\Delta C}_{t^{\prime} }\) (Extended Data Fig. 1b and Supplementary Fig. 11). To determine the sectoral allocation of the overall capacity to the 14 end uses after 2030, we used the green hydrogen end-use shares of the IEA NZE Scenario⁴⁰ (Supplementary Fig. 13). The results for this 1.5 °C scenario until 2050 are presented in Supplementary Fig. 16.

Levelized costs of green products

For all green products, we first calculated LCOH for each year from 2024 using the annuity method and broadly following the system boundaries outlined in ref. ⁶² (for the parameters, see Extended Data Table 2), but adding end-use-specific transport and storage costs (Supplementary Table 2). Omitting time indices, the LCOH was calculated as:

where \({\eta }_{{\rm{H}}_2}\) denotes the electrolyser efficiency, \(a(r,\tau )=\frac{r}{1-{(1+r)}^{-\tau }}\) is the annuity factor, \(r\) is the cost of capital, τ is the payback period in years (which can be shorter than the technical lifetime), τ_stack is the lifetime of the electrolyser stack in years, \({{\rm{FOM}}}_{{\rm{H}}_2}\) is the fixed operation and maintenance costs as a percentage of the specific investment costs, I_BOP is the specific investment cost of the electrolyser’s balance of plant (BOP) and other engineering work, I_stack is the specific investment cost of the electrolyser stack, \({{\rm{FLH}}}_{{\rm{H}}_2}\) is the electrolysis full load hours, p_elec is the price of electricity and \({{\rm{VOM}}}_{{\rm{H}}_2}\) is the variable operation and maintenance costs, which are transport and storage costs (Supplementary Table 2). Both I_BOP and I_stack relate to the electrical input capacity of the electrolyser (US$ kW_el⁻¹).

The electricity price paid by electrolysers is highly dependent on the specific supply case and the regulatory definition of green hydrogen with respect to spatio-temporal matching and additionality^28,29. Flexible operation and a direct connection to a renewable energy source reduces the price as electrolysers can tap into hours when electricity is cheap and abundant. Grid-connected electrolysers need to pay grid fees on top of electricity prices, but can run at higher full load hours. Furthermore, stationary batteries can extend the electrolyser’s full load hours by providing a buffer for renewable electricity, but require additional investments. While hourly energy system models can represent these effects in detail²⁸, we accounted for them in an aggregated manner by using the same broad range of electricity prices as in ref. ²⁷. This ensures high traceability of results, while still capturing the effects of system heterogeneity. Further discussion is provided in Supplementary Note 2, while Supplementary Note 3 discusses how energy system models could learn from our results.

We separated the total specific investments costs of the electrolyser (I) into I_stack and I_BOP because (1) the stack needs to be replaced earlier than the rest of the electrolyser, such that we included two annuities in equation (4)⁶², and (2) the stack is much more modular and therefore more susceptible to cost improvements¹⁷, which we included through different learning rates. Technological learning reduces specific investment costs of both I_BOP and I_stack in year t (I_t) according to

where I₂₀₂₃ denotes the investment costs in 2023, C_t denotes the global cumulative electrolysis capacity in year t, C₂₀₂₃ = 0.92 GW installed capacity in 2023 and LR denotes the learning rate. Technological learning is driven by cumulative project announcements until 2030 and subsequently by the median 1.5 °C scenario (Supplementary Fig. 11). Thus, electrolyser costs fall quickly (Supplementary Fig. 12).

For electrofuels derived from green hydrogen (e-kerosene, e-methanol and e-methane), the corresponding LCOX are

where FOM_X represents fixed operation and maintenance costs, I_X is the specific investment cost of the electrofuel synthesis plant (in terms of electrofuel output), FLH_X is the full load hours of the synthesis plant, \({p}_{{\rm{H}}_2}={{\rm{LCOH}}}-{{\rm{VOM}}}_{{\rm{H}}_2}\) is the price of hydrogen (that is, the LCOH without transport and storage costs), η_X is the synthesis energy efficiency, \({p}_{{{\rm{CO}}}_2}\) is the price of renewable CO₂ (not the carbon price of emissions), ε_X is the CO₂ intensity of the electrofuel and VOM_X is the end-use-specific transport and storage costs (Supplementary Table 2).

The price of renewable CO₂, which can either come from biogenic sources or from direct air capture, is an uncertain but important cost component for the production of carbon-neutral electrofuels (Extended Data Fig. 5g–l). While biogenic carbon can initially be as cheap as US$30 tCO₂⁻¹, it likely faces availability limits such that it could quickly become more expensive as demand increases (see, for example, Fig. 6.3 in ref. ⁶³). In contrast, direct air capture is more scalable, but currently faces very high costs in the order of US$500–1,000 tCO₂⁻¹, which could reduce to approximately US$300 tCO₂⁻¹ once the scale of 1 GtCO₂ yr⁻¹ is reached in the long term⁶⁴, although this is again subject to substantial uncertainty. In our central estimate, we set the average cost of renewable carbon to US$200 tCO₂⁻¹, which reflects the different CO₂ sources reported in electrofuel projects, while the progressive and conservative sensitivity scenarios covered a wide range of US$30–300 tCO₂⁻¹.

Prices of fossil competitors

We collected harmonized data on prices for all fossil competitors represented in our pay-as-bid market premium model for 2024, 2030 and 2050 (for parameters, see Extended Data Table 3), using linear interpolation in between. For natural gas, our cost estimate was the average of the EU trading point Title Transfer Facility in the Netherlands and the US trading point Henry Hub, using spot market prices in 2024 and future prices in 2030. For 2050, we used the gas price from the IEA NZE 1.5 °C scenario⁴⁰. For grey hydrogen and grey methanol, which are produced from natural gas, we first collected current prices for 2024. To ensure internal consistency with natural gas prices, we then calculated the corresponding specific fixed costs in 2024, which reflect the per-megawatt hour capital costs associated with the synthesis plant. Assuming that these stay constant, for 2030 and 2050 we inferred the price of grey hydrogen and grey methanol by adding the corresponding variable costs, that is, the natural gas price divided by the efficiency. We proceeded similarly for kerosene and diesel, using crude oil spot and future prices as the reference for 2024 and 2030, respectively, while for 2050 we again used the oil price from the IEA NZE 1.5 °C scenario. This calibration ensured that prices for fossil products are internally consistent.

Last, we differentiated between scenarios without and with ambitious carbon pricing. For the latter, we used a carbon price pathway that is in line with EU climate targets in the sectors covered by the EU Emissions Trading System, such as industry and energy supply⁴¹. The CO₂ cost per megawatt hour of the fossil competitor is the product of the emissions intensity, including upstream methane emissions for natural gas, grey hydrogen and grey methanol²⁷, and the carbon price per tonne of CO₂. We denote the total cost as p^fossil, which includes CO₂ costs if applicable. In addition, for natural gas, we considered grid fees of US$5 MWh⁻¹ based on ref. ⁶⁵ (Supplementary Table 2).

Limitations

As the quality of the data of the IEA Hydrogen Production and Infrastructure Projects Database⁵⁸ may be limited, we conducted a comprehensive data validation (see the section ‘Data quality validation’, Supplementary Note 1, Supplementary Table 1 and Supplementary Figs. 1–4). Nevertheless, some errors may remain, particularly for smaller projects that were not checked. In general, there are counteracting uncertainties related to project announcements. On the one hand, the database may underestimate projects, as we verified only existing entries and did not conduct research to identify potentially missing projects. On the other hand, the database may include projects that are no longer active, as it is often unclear if and when a project has been scrapped.

The quality of the data of the electrolysis requirements in 1.5 °C scenarios is limited due to heterogeneous sources and limited numerical reporting of the scenario data accompanying the reports. In several cases, we had to infer electrolysis capacity from green hydrogen production values, also for IAM scenarios. Thus, Fig. 4 and Extended Data Fig. 1 show only estimates of electrolysis capacity using publicly available data and should not be interpreted as numerically exact.

Modelling the pay-as-bid market premium to estimate subsidies required several simplifications. First, although we distinguished between 14 end-use applications, four green products and five fossil competitors, we did not account for regional differences in hydrogen production costs. Our estimates can be interpreted as cross-regional averages. Note that our sensitivity ranges are large enough to contain the regional cost heterogeneity found in GIS-based analyses⁶⁶. Second, we neglected additional end-use transformation costs, which are typically small or even zero, for example, for drop-in electrofuels. Some applications can simply replace grey with green hydrogen with no additional costs (for example, ammonia production), while additional investment costs in other applications are low compared with fossil applications (for example, direct reduced iron-based steel plants or hydrogen boilers). Third, we calculated levelized costs using constant electricity prices, assuming that green hydrogen projects require new dedicated renewable energy plants or long-term contracted power-purchase agreements that deliver electricity at stable prices. Similarly, for electrofuels, this implies dedicated electrolysers or long-term contracts that deliver green hydrogen at constant prices. Fourth, we did not consider the option that projects could pay back a part of the received subsidies once they are profitable relative to their fossil competitor in the future because this would require a contract for differences that allows for this option. Fifth, we did not include factors other than costs that influence the project realization as this was outside the scope of this analysis. Sixth, we did not incorporate the competition of green hydrogen with blue hydrogen and other mitigation options, which we discuss in Supplementary Note 4. Last, we assumed that demand-side policies directly translate into electrolysis capacity without the need for additional subsidies.

The quality of the data of global announced hydrogen subsidies from BloombergNEF (BNEF) may be limited and will likely soon be outdated. The estimate for US subsidies is particularly uncertain as the production tax credits of the Inflation Reduction Act¹² are uncapped such that BNEF bases their US subsidy estimates on hydrogen project announcements. Furthermore, the tracked subsidies cover not only green hydrogen but also other sources of low-carbon hydrogen, which we optimistically compared to subsidy requirements only for green hydrogen project announcements. The global subsidy volume of US$308 billion for low-carbon hydrogen as of September 2023 therefore serves only as a snapshot. Although this figure will be outdated soon, it still offers a valuable reference point. However, it should be interpreted with caution as the implementation of these subsidies will critically depend on future government commitments to foster the hydrogen market ramp-up.