Sample
As described in the flowchart in Supplementary Fig. 9, we start by describing the sample. We model the LCOH for all African countries with access to ports, excluding landlocked countries, due to the logistical and infrastructural complexities that hinder H2 export from these areas. Further, Somalia and Libya are excluded from our analysis given that in the past 5 years, both countries were in the bottom 5% of the World Bank Governance Indicators in terms of political stability37. It is therefore likely that investors would refrain from any project in these countries, irrespective of the theoretical COC. Finally, we exclude small island states such as Cape Verde or Mauritius from our analysis due to space and infrastructure constraints. This yields a list of 31 African countries for our sample, which constitutes 85% of total African GDP38.
To collect planned green H2 projects, we use the International Energy Agency’s (IEA) Hydrogen Database, which lists 1,991 H2 projects as of December 202322. Of these, 66 projects are located in our sample countries and plan to produce H2 from electrolysis using renewable electricity from either wind or solar. Note that our scope excludes two projects in Zimbabwe, one of which was decommissioned in 2015, and one project in Niger. We further restrict our sample to projects planned to go online by 2030 for two reasons. First, announced projects with live dates beyond 2030 are probably speculative, and it is difficult to assess the credibility of the plans. Second, the COC and several other cost factors, such as the cost of renewables or the cost of NH3 shipping are changing over time, making cost projections beyond 2030 difficult.
The final green H2 project sample consists of 34 projects, for which we include the project’s development status, planned first year of operation, designated end-use applications and size in standardized electrolysis capacity as calculated by the IEA in MW H2 output (LHV) for all Power-to-X projects (Supplementary Table 3)22. We use the median planned capacity of 60.6 kt H2 yr−1 as the green H2 demand for the LCOH modelling. Capacity affects the LCOH via economies of scale (for example, in the electrolyser) and space constraints as larger projects need more space, mainly for renewable energy build-out.
Estimating the cost of capital
Between February and August 2023, F. Schneider and F. Egli conducted 12 virtual exploratory expert interviews with 13 representatives to inform the financing scenarios shown in Table 1. Because green H2 projects at scale are currently hypothetical on the African continent, the interviews served to understand the planned financing structures. All interviews followed the same question guide provided in a slide deck to interviewees. Interviewees were sampled from organizations that would probably be involved in financing deals if de-risked by European policymakers. Early interviewees were contacted using the researchers’ network and subsequent ones via snowball sampling. Systematic sampling is impossible because few experts globally can comment on the planned financing structures. Note that interviews only served to triangulate our COC estimation approach, which is based on peer-reviewed literature. Hence, we are less concerned about potential sampling biases regarding gender or region. Interviews took place under Chatham House rules, and consent to use provided information in research was obtained at the beginning together with shared information on the research project in the form of a slide deck. No personal information beyond participants’ names and affiliations was obtained and interviewees are listed anonymously only. An overview of the interview sample is provided in Supplementary Table 4.
The COC is the price that a profit-maximizing capital provider demands for investing equity into a project or issuing debt (for example, loans) for a project. The COC increases with the risk for an investor of being unable to recoup their investment, for example, due to uncertain policy environments or novel risky technologies. In financial economics, it is common practice to decompose the COC into a risk-free rate (reflecting the time value of money) and a risk premium (reflecting the investment-specific risk). The latter typically differs between countries, technologies and over time39. A standard project-level specification of the COC is the weighted average cost of capital (WACC), where capital is sourced from equity and debt financing. The WACC reflects the costs of obtaining debt and equity financing, respectively, and the share of each type within the total capital budget. In line with the literature39, a standard notation ‘vanilla-WACC’ (no consideration of potential tax deductions for debt payments) can be defined as follows:
$$\mathrm{WACC}=\left(\frac{E}{V}\times {K}_{\mathrm{e},i}\right)+\left(\frac{D}{V}\times {K}_{\mathrm{d},i}\right)$$
(1)
where \({K}_{{\mathrm{e}},{{i}}}\) and \({K}_{{\mathrm{d}},i}\) denote the cost of equity and the cost of debt, respectively, for investments in a specific country i. \(E\), \(D\) and \(V\) denote total equity, debt and capital; the debt share is denoted as \(\frac{D}{V}\). As we model the case of exporting green H2 from Africa to Europe, it is uncertain which entities would be liable to pay tax where and we do not consider a country-specific tax rate. We use the terms COC and WACC interchangeably in this paper, focusing on COC in the main text for simplicity. In the absence of a track record for the financing of green H2 projects globally and certainly in Africa, we define four financing scenarios to model the COC based on insights from the finance literature and expert interviews (Table 1, main text).
Across all financing scenarios, we use a separate COC for the plant investment encompassing H2 production facilities (for example, the electrolyser), the renewable energy generation assets and the supporting infrastructure encompassing roads, pipelines and so on. The risk-free rate \({r}_\mathrm{f}\) is based on two indicators: a long-term risk-free bond, commonly depicted with the 10-year US treasury bond yield and an overnight interbank rate reflecting the current interest rate environment, commonly depicted with the Effective Federal Funds Rate (FFR). In line with previous work40, we estimate \({r}_\mathrm{f}\) for a high and a low interest rate scenario to account for the fact that the interest rate environment has a large impact on the cost of renewables. We set \({r}_{\mathrm{f}_{\mathrm{low}}}\) to 2%, which is reflective of the 5-year average of the 10-year treasury bond in the aftermath of the financial crisis of 2008 between 2009 and 2013 (2.68%), considering the FFR was substantially lower during that period (0.14%)41,42. Conversely, \({r}_{\mathrm{f}_{\mathrm{high}}}\) is set to 5%, which is representative of the high interest environment over the last year (August 2023–July 2024) at the time of writing. During this period, the average 10-year US treasury bond yield stood at 4.33% and the FFR at 5.33% (refs. 41,42). In this Article, we therefore consider the high interest rate environment (Table 1) to be representative of the status quo. The share of debt in total financing, \(\frac{D}{V}\), is assumed to be 75% across all scenarios18.
We model a commercial scenario in both interest rate environments (Table 1, main text). For these commercial scenarios 1 and 3, we define the cost of debt to reflect lending to a large infrastructure project in a specific country. Namely, we add a country default spread to reflect country risk43 (\({\mathrm{CDS}}_{{\mathrm{Host}}_{i}}\)) and a lender margin (\({L}_\mathrm{m}\)), which we set to 2% in line with the literature15,44,45, to reflect infrastructure risk. The country default spreads are reflective of country risk at the time of writing in 2023. The cost of debt for the plant is therefore given by:
$${K}_{\mathrm{d},i,\mathrm{commercial},\mathrm{plant}}={r}_{\mathrm{f}_{\mathrm{low},\mathrm{high}}}+{\mathrm{CDS}}_{{\mathrm{Host}}_{i}}+\mathrm{Tp}$$
(2)
Similar to the cost of debt, the cost of equity contains a country mark-up. Furthermore, we add an equity risk premium and a technology premium to reflect the additional risk of equity compared to debt and the risk of green H2 investments, as there is a very limited track record. The cost of equity for commercial scenarios was calculated as follows:
$${K}_{\mathrm{e},i,\mathrm{commercial},\mathrm{plant}}={r}_{\mathrm{f}_{\mathrm{low},\mathrm{high}}}+\mathrm{ERP}+{\mathrm{CRP}}_{{\mathrm{Host}}_{i}}+\mathrm{Tp}$$
(3)
where ERP is the equity risk premium of a mature market, set to 5% in July 202343. \({\mathrm{CRP}}_{{\mathrm{Host}}_{i}}\) varies by country i and accounts for the return that investors require as compensation for the risk of an investment in a publicly listed company in each country. In addition, the technology premium (Tp) reflects that green H2 is a relatively immature technology with a limited track record of successfully constructing large-scale projects. Following a recent IRENA report15, Tp is set to 3.25%, reflecting an investment premium for novel technologies. Because ref. 43 does not provide \({\mathrm{CDS}}_{{\mathrm{Host}}_{i}}\) and \({\mathrm{CRP}}_{{\mathrm{Host}}_{i}}\) for Eritrea, Equatorial Guinea, Djibouti and Mauritania, \({\mathrm{CDS}}_{{\mathrm{Host}}_{i}}\) is obtained using Wikiratings as described in Supplementary Table 5. Thereafter, \({\mathrm{CRP}}_{{\mathrm{Host}}_{i}}\) is calculated following the approach suggested by ref. 43.
Finally, we assume that any supporting infrastructure will be financed by a project’s host government at its sovereign rate. Consequently, the COC for infrastructure is given by:
$${\mathrm{COC}}_{i,\mathrm{commercial},\mathrm{infra}}={r}_{\mathrm{f}_{\mathrm{low},\mathrm{high}}}+{\mathrm{CDS}}_{\mathrm{Host}}$$
(4)
For the de-risked scenarios 2 and 4 (Table 1, main text), we model a situation where a green H2 project on the African continent benefits from access to below-market terms financing due to an offtake guarantee from a western European government entity. This assumption follows developments driven in particular by Germany, which has established diplomatic relations to support the transition of current fossil fuel exporting nations such as Angola or Nigeria to a decarbonized energy export industry by substituting fossil fuel exports at least partially by H2 (ref. 46). Moreover, Germany has recently announced a joint declaration of intent with the Netherlands to implement a joint tender under the H2Global Instrument, offering 10-year purchase agreements to suppliers to kick-start the emergent European green H2 import market47. Finally, Germany has signed further bilateral partnership agreements with countries such as South Africa48, Namibia49 and Kenya50.
In these scenarios, the cost of debt can be represented as follows:
$${K}_{\mathrm{d},i,\mathrm{derisked}}={r}_{\mathrm{f}_{\mathrm{low},\mathrm{high}}}+{\mathrm{CDS}}_{\mathrm{WesternEU}}+{\mathrm{MIGA}}_{{\mathrm{expr}}_{i}}+{\mathrm{MIGA}}_{{\mathrm{war}}_{i}}$$
(5)
where \({\mathrm{CDS}}_{\mathrm{WesternEU}}\) represents the average default spread of a western European country weighted by its GDP, where western Europe includes Andorra, Austria, Belgium, Cyprus, Denmark, Finland, France, Germany, Greece, Guernsey, Iceland, Ireland, Isle of Man, Italy, Jersey, Liechtenstein, Luxembourg, Malta, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, Turkey and the United Kingdom43. In July 2023, this amounted to 0.96% (ref. 43). Despite the offtake guarantee, certain risks, such as the risk of expropriation or war, will remain. Consequently, we assume that in scenarios 2 and 4, investors will seek insurance against such political risks, which could disrupt operations or damage assets. Informed by the expert interviews and because private political risk insurance is not available in most countries in our sample, we assume political risk insurance by the World Bank Group’s Multilateral Investment Guarantee Agency (MIGA). \({\mathrm{MIGA}}_{{\mathrm{expr}}_{i}}\) and \({\mathrm{MIGA}}_{{\mathrm{war}}_{i}}\) represent the price for obtaining such coverage for war and expropriation risk. As the MIGA pricing is confidential, we develop a heuristic to approximate the pricing based on reports and the expert interviews. Reference 51 states that the price per MIGA risk ranges from 0.5% to 1.75% of the total sum insured, depending on the country and project risk. Assuming that in a de-risked scenario, only country risk will remain as the project is fully de-risked, the distribution of in-sample country risk, reflected by the credit default spread provided by ref. 43 can be mapped onto the pricing range indicated by ref. 51. Formally, the approach can be represented as follows:
$${\mathrm{MIGA}}\,{\mathrm{Risk}}\,{\mathrm{Pricing}}\left(x\right)=f\left(g\left(x\right)\right)$$
(6)
where \(x\) is the percentile of the default spread of a country based on ref. 43, \(g\left(x\right)\) returns the percentile of the CDS in the sample distribution and f() maps the percentile to the corresponding percentile of the MIGA pricing range51.
The cost of equity was calculated as follows:
$${K}_{\mathrm{e},i,\mathrm{derisked},\mathrm{plant}}={r}_{\mathrm{f}_{\mathrm{low},\mathrm{high}}}+\mathrm{ERP}+{\mathrm{CRP}}_{\mathrm{WesternEU}}$$
(7)
where \({\mathrm{CRP}}_{\mathrm{WesternEU}}\) reflects the average equity country risk premium in Western Europe weighted by GDP. In July 2023, this premium was 1.37% (ref. 43).
Finally, we assume that infrastructure in the de-risked scenarios is either financed by the host government, that is, as in the commercial scenarios or financed by the project sponsor backed with an offtake guarantee from a western European government. We therefore define the COC for infrastructure investments in the de-risked scenarios as the minimum of the host government’s sovereign rate \({\mathrm{WACC}}_{\mathrm{commercial},\mathrm{infra}}\) and the de-risked COC based on \({K}_{\mathrm{e},i,\mathrm{derisked},\mathrm{plant}}\) and \({K}_{\mathrm{d},i,\mathrm{derisked},\mathrm{plant}}\). A detailed breakdown of how the COC components were obtained and the corresponding data sources are provided in Supplementary Table 5.
Modelling the LCOH
The GeoH2 optimization model is used to calculate the lowest possible cost of H2 achievable throughout each country, assuming an electrolyser lifetime of 20 years. Electrolyser lifetimes are subject to some uncertainty, but do not have a major impact on LCOH52 and most importantly for this analysis, affect the LCOH in European and African countries alike except for a small penalty for Africa-based production due to higher financing costs on the CAPEX. The model tessellates the country into hexagons and calculates the costs to (1) produce the specified quantity of green H2 (or here green NH3) in each hexagon, (2) convert it to the required state for transport and (3) transport it to a specified demand location. In each hexagon, a cost-optimal off-grid H2 plant powered by PV and wind turbines is designed to meet the specified demand. The electrical infrastructure (that is, PV, turbines, battery storage) and plant infrastructure (that is, electrolyser, NH3 storage, compressed H2 storage) are sized for cost optimality using site-specific, hourly weather data from the European Centre for Medium-Range Weather Forecasts Reanalysis v5 dataset53. Here data for the duration of 2022 are used. The Corine Land Cover54 and OpenStreetMap55 datasets are used to constrain land availability in each hexagon, and country boundary base maps are from GADM, which allows free academic re-use. The costs to transport the H2 to port are calculated for both road transport (that is, trucking) and pipeline transport, including construction of necessary infrastructure. Water costs for either desalination or freshwater processing are included as applicable—however, no limit is placed on water consumption to avoid depletion in either case. Cost parameters used in the modelling are available in Supplementary Table 1 and harmonized to 2023 euros using average annual US$/€ exchange rates from the Economic Research Division of the Federal Reserve Bank of St. Louis and an annual average of the Harmonized Index of Consumer Prices from Eurostat. Further details on the GeoH2 model implementation are available in the model descriptor14.
The model is applied to each country in the project sample. A demand of 60.6 ktH2 yr−1 is simulated at each country’s main port. This demand is assumed to be met in the form of green NH3 (that is, 341.4 ktNH3 yr−1) due to its cost advantages in shipping and to be temporally uniform (that is, evenly spaced truck pick-ups throughout the year or a consistent pipeline flow rate). As NH3 is utilized as a transport vector, we build on the ammonia trade literature. A first stream of this literature investigates green ammonia costs in/from specific countries56,57 and trading routes58 whereby a second stream of literature looks at global ammonia trade. Reference 59 applies a global optimization model to locate optimal ammonia production sites, however the site selection is based on a previous paper and planned projects, considering a total of 112 sites. Such models are great to assess the cost competitiveness of planned projects; however, they cannot compare these to alternative possibilities as we do for the African continent. Finally, ref. 60 applies a geospatial model in 28 countries with a less granular resolution within country and abstracting from water costs and transportation options to port (pipeline vs trucking), which are included in this paper. Hence, whereas the key contribution of this paper is the inclusion and comparison of different financing and policy scenarios and the explicit calculation of the cost differentials for the entire African continent, we also apply a state-of-the-art bottom-up model, which can serve to improve within-country cost comparisons.
Country-specific figures are used for energy prices, heat prices and interest rates. Level-four H3 hexagons61 are used to define the spatial resolution. Land availability is constrained such that H2 production and associated generation are not permitted to be built on wetland, built-up areas, water bodies or within 250 m of coastlines or protected areas. PV is additionally not permitted to be built on agricultural land. Whereas elevation is not considered as an exclusion criterion here due to data constraints, future work may also wish to exclude high elevations or steep slopes. All maps are based on the authors’ own analysis of publicly available input data. Note that this work leverages a model of the Haber–Bosch process in plant optimization in place of the H2 production process available in the standard GeoH2 model14. This variant of the codebase is made available on GitHub (Code Availability).
To account for shipping costs, the sea distance from each of the exporting ports to Rotterdam is first calculated using the ShipTraffic website62. Previous work has estimated the cost of shipping NH3 over a distance of approximately 13,800 km to be €0.39 kgH2−1 (ref. 13). Following ref. 24, shipping cost projections depend approximately linearly on transport distance. Consequently, we scale this estimate linearly to km, resulting in our cost parameter of €0.00003 kgNH3−1 km−1, which we multiplied with each of the obtained distances from the African port to Rotterdam. Implementing this approach yields a shipping cost range of €0.09 kgH2−1 (Morocco)–€0.44 kgH2−1 (Mozambique), in line with other estimates in the literature, according to which shipping could add up to €0.46 kgH2−1 by 203063.
The interest rates for Rotterdam are obtained following the same approach as for all other countries described in Methods and following previous work, heat costs of €0.06 kWh−1 are assumed for converting NH3 to H2 (ref. 13). Note that the cost of this process is subject to some uncertainty because it does not yet exist at scale. Electricity costs are assumed to be €0.1 kWh−1 and are calculated as the combination of the average price of Dutch Power Base futures64 and the price of a guarantee of origin for renewable electricity. At the time of writing, Dutch Power Base futures are available until October 2028, and the average price obtained is €0.097 kWh−1. Ideally, we would base our calculation on a Dutch Power Base future that matches our modelling period of 20 years. As this is not available, we took the longest available market-based future as input for our simplified model in Rotterdam using constant electricity prices. Moreover, future prices reflect the current interest rate environment; hence, we deem our Rotterdam costing most representative for scenarios 1 and 2, on which all results in the main text are based and suggest some caution when using it for scenarios 3 and 4. On the basis of grey literature65, an average price of a guarantee of origin of €0.055 kWh−1 by 2030 is assumed. Whereas Rotterdam serves as our comparison case, the resulting electricity cost is deemed representative of the European Union as a whole, given that historically, Dutch wholesale electricity prices were strongly correlated with German wholesale electricity prices, and the Dutch wholesale price roughly represents the average wholesale electricity price in Europe66.
Our modelling excludes two cost components: namely, (1) costs associated with upgrading ports to enable large-scale NH3 shipments and (2) costs for last-mile distribution in Europe. Both would require detailed information (that is, on port design and demand locations respectively), which is beyond the scope of this analysis. Furthermore, our model does not account for potential cost reductions in onshore wind, solar PV, electrolysers and battery storage that may occur by the year 2030. Such cost decreases will not only reduce the cost of green H2 projects in Africa but may also influence renewable deployment in Europe and, therefore, European wholesale electricity prices. As such, the net effect on the cost competitiveness of African green H2 exports vs European green H2 production remains inconclusive.
Finally, we calculate the LCOH in Rotterdam to create a European cost benchmark. We use the same assumptions for electricity and heat costs as mentioned above to model LCOH for green H2 produced in Rotterdam using grid electricity due to space constraints for renewable energy. Due to the absence of renewable energy investments, these projects are much less capital intensive, and variations in the COC, therefore, are less important for LCOH. Note that recent auctions by the European Hydrogen Bank have yielded winning projects using grid electricity and/or greenfield renewable energy. Here we consider the former case only, whereas future research could model European least costs more comprehensively by considering all locations in Europe with their respective transport costs to demand centres. We calculate an LCOH for production in Rotterdam for each financing scenario shown in Table 1 and obtain an LCOH of €4.74 kgH2−1 for scenario 1, €4.72 kgH2−1 for scenario 2, €4.69 kgH2−1 for scenario 3 and €4.67 kgH2−1 for scenario 4 (Supplementary Table 2). Because large green H2 production plants do not currently exist, cost estimates are not commonly available, but several reports have tried to estimate costs. These are broadly in line with our costs; for example, Aurora Energy Research estimates the least-cost LCOH in Germany by 2030 between €3.9 and €5 kgH2−1 (ref. 67). Other research reports even lower 2030 costs for Germany of US$3.1 kgH2−1 in a baseline scenario and US$2.7 kgH2−1 in an optimistic scenario68, which is roughly the range where the IEA Global Hydrogen Review places North-Western European green H2 costs by 2020 (€3.1 kgH2−1) (ref. 1). Other European locations, such as Spain, with more favourable renewable energy sources and similarly favourable financing costs, may reach even lower costs by 2030 at €2.7 kgH2−1 as estimated by the Hydrogen Council and McKinsey69. These costs have been confirmed by the results of the recent European Hydrogen Bank auction yielding a lowest bid in Spain at €2.8 kgH2−1. For the four other countries where least costs were disclosed, they ranged from €4.6 kgH2−1 in Norway to €7.6 kgH2−1 in the Netherlands (Supplementary Table 6). Note that we prefer comparing our results to realized costs as comparisons across studies with different methodological approaches, spatial resolutions, geographical and technological scopes and policy cases considered are difficult (Supplementary Table 7). A thorough comparison of different studies on the subject would require a separate review article with a methodological approach to render comparisons meaningful.
