Quantifying the cost savings of global solar photovoltaic supply chains

Achieving carbon neutrality requires deploying renewable energy at unprecedented speed and scale 1,2 , yet countries sometimes implement policies that increase costs by restricting the free flow of capital, talent and innovation in favour of localizing benefits such as economic growth, employment and trade surpluses 3,4 . Here we assess the cost savings from a globalized solar photovoltaic (PV) module supply chain. We develop a two-factor learning model using historical capacity, component and input material price data of solar PV deployment in the United States, Germany and China. We estimate that the globalized PV module market has saved PV installers US$24 (19–31) billion in the United States, US$7 (5–9) billion in Germany and US$36 (26–45) billion in China from 2008 to 2020 compared with a counterfactual scenario in which domestic manufacturers supply an increasing proportion of installed capacities over a ten-year period. Projecting the same scenario forwards from 2020 results in estimated solar module prices that are approximately 20–30 per cent higher in 2030 compared with a future with globalized supply chains. International climate policy benefits from a globalized low-carbon value chain 4 , and these results point to the need for complementary policies to mitigate welfare distribution effects and potential impacts on technological crowding out.

Achieving carbon neutrality requires deploying renewable energy at unprecedented speed and scale 1,2 , yet countries sometimes implement policies that increase costs by restricting the free flow of capital, talent and innovation in favour of localizing benefits such as economic growth, employment and trade surpluses 3,4 . Here we assess the cost savings from a globalized solar photovoltaic (PV) module supply chain. We develop a two-factor learning model using historical capacity, component and input material price data of solar PV deployment in the United States, Germany and China. We estimate that the globalized PV module market has saved PV installers US$24 (19-31) billion in the United States, US$7 (5-9) billion in Germany and US$36 (26)(27)(28)(29)(30)(31)(32)(33)(34)(35)(36)(37)(38)(39)(40)(41)(42)(43)(44)(45) billion in China from 2008 to 2020 compared with a counterfactual scenario in which domestic manufacturers supply an increasing proportion of installed capacities over a ten-year period. Projecting the same scenario forwards from 2020 results in estimated solar module prices that are approximately 20-30 per cent higher in 2030 compared with a future with globalized supply chains. International climate policy benefits from a globalized low-carbon value chain 4 , and these results point to the need for complementary policies to mitigate welfare distribution effects and potential impacts on technological crowding out. Solar energy is promised to play a crucial role in achieving a sustainable, low-carbon energy future and avoiding the worst impacts of climate change 1 . Over the past 40 years, solar photovoltaic (PV) prices have fallen by over two orders of magnitude, and during the period 2010 to 2021, the global weighted-average levelized cost of energy of newly commissioned utility-scale solar PVs fell by 88% (ref. 5 ), making solar PVs cheaper than fossil fuel power in some parts of the world. Installed costs (excluding the cost of capital) fell by 81% over this period 5 . Although these dramatic price declines have been a boon for accelerating low-carbon energy deployment 6 , further declines will be necessary to deploy renewables at the speed and scale that is needed to achieve climate targets, especially in the remaining parts of the world where fossil fuel power is still cheaper 7 . Recent research suggests that the rates of solar and wind energy deployment in even the fastest-deploying nations are not high enough to meet the targets necessary to avoid the worst consequences of climate change 8 .
Nonetheless, rapid price declines in solar PV have not been without controversy. China, for example, has played an outsized role in scaling up the mass production of solar PV cells and modules, comprising 78% of global production in 2021 9,10 ( Fig. 1). Greg Nemet went as far as to call this outcome China's "gift to the world" 11 , referring to the dramatic manufacturing cost reductions achieved by Chinese firms in the past decade 5 . Yet other nations view the concentration of PV manufacturing in China as a competitive threat, and some have attributed this outcome to unfair trade practices and industrial policies implemented by China's government 12 . Countries seeking to capitalize on the growing clean energy sector are looking to protect and grow domestic manufacturers 3 .
In response to these concerns, the United States and the European Union have imposed steep solar tariffs on imports from China and other countries. In June 2022, the Biden administration invoked the Defense Production Act to accelerate the onshoring of solar PV manufacturing 13 . These efforts could lead to less efficient national learning processes replacing the learning processes associated with global supply chains that have led to drastic price declines 4 . The free flow of capital (for example, foreign finance-backed start-ups), talent (for example, international collaborations with Chinese researchers) and innovations (for example, technologies pioneered in labs overseas and licensed and mass-produced in China) were essential to the rise of China's competitive solar PV industry 14 . Each of these activities is increasingly under scrutiny by the United States and other governments 15 . In the event of strict nationalization policies (including, inter alia, trade barriers in final or intermediate solar goods, restrictions on cross-national research and development, and barriers to cross-border investment), subsequent cost and performance improvements could derive primarily from activities, knowledge and capital within national borders, potentially slowing the rate of price declines in globally traded solar PV components and, consequently, the rate of solar PV deployment.
International climate policy and renewable energy deployment policy now face a crossroads: continue relying on global supply chains, or pivot towards domestic technology development and production.

Article
This study attempts to quantify the difference between these two paths in terms of the costs of deploying solar PV to achieve ambitious low-carbon goals. We collect detailed historical capacity, component and input material cost data of solar PV deployment in the United States, Germany and China, and develop a two-factor learning model to estimate a learning curve associated with the historical (globalized) solar PV supply chain. We then use these learning models to compare counterfactual historical prices and potential future prices of solar PV modules under 'global' versus 'national' market conditions. The global market scenarios reflect learning under historical market conditions whereas the national market scenarios reflect a gradual transition to fully domestically supplied markets over a ten-year period in each country.
We focus our scope on PV modules for two reasons. First, modules are a globally traded component and comprise between 20% and 40% of the installed system cost for most PV installations 16 ; combined with inverters, modules comprised 61% of the global weighted-average total installed price decline between 2010 and 2020 5 (although they are expected to account for lower portions of cost in the future). Second, other 'soft costs' (such as permitting, installation and marketing) vary widely by country and have geographically limited learning and spillover effects 17 ; as a result, we expect these cost components to remain relatively similar regardless of where modules are manufactured. Our analysis is limited to installed prices, not the levelized cost of energy as reflected in power purchase prices for solar energy, which also vary by country and project according to the cost of capital and other factors.

Modelling historical prices and savings
Using nation-specific, component-level price data and global PV installation and silicon price data, we estimate learning rates for solar PV modules in the three largest solar-deploying countries (China, Germany and the United States) between 2006 and 2020 using a two-factor learning model. Combined, these three markets comprised 54% of all global installed PV capacity during this period 18 . Estimated learning rates during this period are 20% in Germany, 26% in the United States and 33% in China. We then compute the counterfactual 'national markets' scenario by assuming that starting in 2006 countries began implementing nationalistic policies that gradually restrict learning to installations within their country borders over a ten-year period (for China, the starting year is 2007 owing to data availability). Annual installed capacities are assumed unchanged in the counterfactual 'national markets' scenario to provide the most policy-relevant results (see more discussion in 'Limitations' in Methods). Figure 2 shows the resulting price curves between the 'global market' and 'national market' scenarios in each country as well as the true historical prices.
Comparing the two scenarios, if each country had pursued a gradual transition to strict nationalistic policies while installing at the same rate over a ten-year period, our results imply that solar PV module prices in 2020 would have been substantially higher than their actual historical prices: 54% higher in China (US$387 per kW versus US$250 per kW), 83% in higher Germany (US$652 per kW versus US$357 per kW) and 107% higher in the United States (US$877 per kW versus US$424   Points are historical module prices, and the two solid lines reflect the modelled prices using global (blue) versus national (orange) markets scenarios. In each modelled curve, the learning rates are held constant by country and silicon prices follow historical global trends (Extended Data Fig. 6). The global market scenario uses global capacities and the national market scenario uses a weighted sum of national and global capacities that reflects a gradual transition to fully domestically supplied markets over a ten-year period. Uncertainty bands represent 95% confidence intervals from the estimated learning models, which were computed via simulation.
per kW). Early learning, boosted in part by Germany's generous solar feed-in-tariffs, led to compounded improvements over time for the United States and China, which led to steep increases in installations in the second half of the period. The combined estimated cumulative savings across all three countries during this period from global versus national markets is US$67 billion (2020 $US), with a 95% confidence interval of US$50-84 billion (Fig. 3).

Future trajectories
As more countries introduce policies aimed at protecting local manufacturers, such as import tariffs on PV modules, continued learning-based reductions in module prices may be delayed. To assess this effect, we project solar PV module prices out to 2030 based on continued global versus national market scenarios starting from historical 2020 PV prices. These projections assume that capacity grows at a constant annual growth rate (CAGR) from 2020 installed capacity levels out to 2030 targets for each country. We consider two different future scenarios: 'national trends', which projects recent deployment trends out to 2030, and 'sustainable development', which reflects more aggressive installation growth to meet climate targets based on the Sustainable Development Scenario in the International Energy Agency World Energy Outlook 2020 7 . Table 1 summarizes the specific 2030 targets for each country in each scenario, and Fig. 4 shows the results of these projections.
These projections imply that prices would be substantially higher in 2030 if strict nationalistic policies were gradually implemented in each country from 2020 to 2030 19 . Therefore, the differences attributed just to domestic production are up to half of the gap between worst-case and baseline cost scenarios. On the basis of the projected installed capacities, the estimated cumulative future savings from 2020 to 2030 across all three countries from global versus national markets is US$15 billion (2020 $US) with a 95% confidence interval of US$13-16 billion under the national trends scenario, and US$29 (27-32) billion under the sustainable development scenario (Extended Data Fig. 1).

Discussion
The manufacturing of solar PV modules-a globally traded commodity that is crucial to addressing climate change-is increasingly contested by governments seeking to localize benefits of the current and future scale of the industry. Yet achieving the rapid rates of solar PV deployment required to address climate change will necessarily require continued price declines at the same or greater rates as those experienced during the past decade, a period during which the free flow of global talent, capital and innovations were instrumental to cost reductions. In this paper, we contribute to understanding the implications of strict nationalistic policies by assembling component-specific solar PV price data (Extended Data Figs. 2-4) across major markets, establishing national-level estimates of learning rates that incorporate silicon prices (Extended Data Fig. 6), and quantifying the potential impact of restricted national learning on historical and projected prices and savings from solar PV deployment. The results may extend to other low-carbon technology sectors, such as wind-generating systems and electric vehicles, with caveats related to the supply chain integration and complexity of technological components. Wind-generating systems, for example, have a very globally integrated and specialized trade in intermediate components 20 ; as a result, achieving 'national markets' for the entire wind supply chain could lead to even larger disruptions in terms of costs and delayed learning.
We identify three dilemmas facing policymakers in preserving established globalized supply chains: trade disputes and domestic employment, 'crowding out' of alternative technology pathways, and additional benefits and drivers of domestic sourcing. Resolving these through complementary policies that mitigate impacts on global learning are difficult but important tasks moving forwards.

Trade disputes and domestic employment
Some have attributed the concentration of PV manufacturing in China to unfair trade practices and industrial policies implemented by China's government 12 . Although constant cost multipliers would be absorbed in the national learning rates, we do not attempt to disaggregate the contributions to these rates nor do we account for changes in national-level producer subsidies or tariffs faced by importers. The 'learning curve' is a synthetic indicator that captures the cumulative effect of impacting factors on the cost evolution of a technology. Data limitations of time-varying government subsidies, industrial policies, tariffs and firm relocations prevent us from disaggregating these precise effects on price and are beyond the scope of this study.
The loss of potential manufacturing jobs in importing countries coupled with trade disputes is prompting much of the impetus for nationalistic policies. The National Renewable Energy Laboratory  Article estimates that there are ten times more annual jobs in system installation compared with those in the entire manufacturing supply chain (although within manufacturing, solar module production is the most labour intensive per gigawatt) 21 . Hence, if higher prices associated with nationalistic policies result in less deployment, total employment may decline, although there may be other redistributive concerns and political realities shaping preferences for certain types of jobs 22 .
Our national markets counterfactual scenario is an illustrative example of more extreme decoupling, although because of the difficulty of onshoring, countries may instead opt to 'near-shore' production to a subset of countries or onshore only select parts of the supply chain. Even the three countries studied could not costlessly onshore entire supply chains; hence, our results probably represent an underestimate of the future costs of strict onshoring policies. Reciprocity in trade policies is another barrier limiting the extent to which nations can fulfil onshoring policy goals: for example, the US polysilicon industry was once a dominant global supplier to solar PV manufacturers but became the first casualty of the solar trade war between China and the United States when China retaliated for tariffs on imported Chinese modules.

Technological 'crowding out'
Some have argued that the rapid price declines of monocrystalline silicon (c-Si) PV cells, driven in part by Chinese industrial policies to ramp up production in China, might have 'crowded out' other emerging solar technologies, such as 'thin film' solar cells for which the United States has a sizable global market share and that could have achieved even lower prices without fierce competition from c-Si 23,24 . Such an argument is not without precedent. For example, ref. 25 found that offshoring manufacturing in the optoelectronics industry to developing East Asia led to such notable price reductions in the incumbent technology that emerging and potentially groundbreaking technologies could not compete and were largely abandoned.
Although these concerns are not without merit, they are not necessarily the only forces at play in the global PV industry. Indeed, PV cell and module manufacturing has followed a developmental path common to many industries in which initial, intense experimentation is followed by the emergence of a 'dominant design' 26 and a shift in productive activity away from product innovations and towards production improvements to increase scale and reduce costs [27][28][29][30] . This shift in focus towards production tends to precipitate two related phenomena: (1) unit costs drop dramatically as firms identify successful production innovations, and (2) many competing firms fail as production tends to concentrate around the handful of firms that are able to compete on lower costs. In some industries, this also coincides with offshoring production in search of lower-cost production environments, although this is not always the case 3 . Thus, it remains unclear whether the concentration of PV cell and module production in China was purely a result of government intervention or perhaps a combination of factors, such as the natural evolution of a maturing industry 31 . Chinese policies may have accelerated cost declines in c-Si cells and modules, but whether they alone led to the crowding out of other potential technologies remains debatable.

Additional domestic and diversified sourcing drivers
A domestic manufacturing base in solar PV may provide other benefits besides direct employment worthy of future study. Our model does not incorporate any spillover benefits to adjacent industries, such as semiconductors and electronics. For example, polysilicon production is part of both advanced chip and solar supply chains, although solar-grade polysilicon has purity requirements several orders of magnitude lower 32 . Establishing a stronger link between public funding of research and development and the private sector has been identified as important to achieving climate technology innovation goals, both by reducing the risks of scale-up and by providing access to markets 33 . Foreign manufacturers may be undesirable or infeasible partners with public money. However, private sector-led efforts can be effective internationally: Chinese solar firms largely innovated through improved manufacturing processes and strategic international partnerships, including with US-based start-ups unable to scale domestically 34 .
Reliance on a single or small set of countries in crucial supply chain bottlenecks, even if reducing costs and enhancing learning, may

Fig. 4 | Comparison of projected solar PV module prices (2020-2030) using global versus national market scenarios in China, Germany and the United
States. Projections assume CAGRs in PV installations to achieve national and global 2030 installation targets. Each curve starts at historical 2020 module prices and follows a nation-specific learning rate. In the global market scenarios, global projected installed capacities are used to project prices whereas in the national market scenarios a weighted sum of national and global capacities is used that reflects a gradual transition to fully domestically supplied markets over a ten-year period. Uncertainty bands represent 95% confidence intervals from the estimated learning models, which were computed via simulation. generate risks of disruption based on natural disasters or geopolitical conflict. A managed diversification-instead of national onshoring-could provide a pathway to mitigate the cost impacts of hardening supply chains.
Finally, maintaining adequate environmental, health and labour standards in the production of traded goods is important for ethical reasons and is increasingly raised in the context of maintaining a level-playing field in trade agreements. The Xinjiang region of China, where much of the world's solar-grade polysilicon is produced, has come under increased scrutiny owing to allegations of forced labour. The solar industry has responded with proposed traceability protocols, which if effective could obviate the need to onshore production for ethical reasons 35 . Further work is needed on the feasibility of such protocols.
This study presents a quantitative estimation of the historical and future cost savings from a globalized solar PV supply chain. The results provide evidence of the benefits of global learning processes in terms of achieving lower prices to accelerate low-carbon technology deployment, which could potentially be delayed by emerging nationalistic policy efforts. When negotiators meet to discuss accelerating action towards the goals of the Paris Agreement, and when policymakers plan for pathways to achieve mid-century carbon neutrality, they should recognize that these aspirations may be difficult or impossible to achieve without globalized low-carbon supply chains. Complementary policies are necessary to address dilemmas and debates with respect to localizing manufacturing and to ensure continued price declines.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-05316-6.

Learning models and simulations
The learning curve model is widely used to describe the evolution of production costs for technologies as they scale up [36][37][38][39][40] . In its simplest form, the learning curve defines a relationship in log-log space between cost (or price) and cumulative capacity 41 . The model can be expanded to incorporate not only the processes of 'learning by doing' but also 'learning by researching' and changes in material input prices 42,43 . Here we adopt a two-factor learning model relating the unit price in year t and country i of solar PV modules, p it , to the cumulative installed PV capacity in year t, q t , and globally averaged polysilicon prices in year t, s t (the primary input material to PV modules): p α β q γ s ln = ln + ln + ln (1) Here, α i is a constant related to starting year conditions in country i, γ i measures the sensitivity to polysilicon prices, and β i is the learning coefficient in country i, which is related to the learning rate (L i ) by: For each country, i, we estimate learning coefficients (Extended Data  Table 1), β i , under historical 'global market' conditions using linear least-squares regression on equation (1). These learning models set a baseline for learning rates under historical market conditions and assume that variations in country-level module pricing were due to transportation, administrative and other non-learning costs.
We then construct counterfactual 'national market' scenarios by assuming that the learning-related price decreases in country i from the starting year, t 0 , are derived from incrementally more nationally installed PV capacity: where q it is the cumulative installed capacity in country i in year t, q jt is the cumulative installed capacity in all other countries in year t, and λ t is a value ranging from 0 to 1. This defines a scenario whereby incremental capacity installed in each year increasingly comes from national as opposed to global installations as λ t shifts from 0 to 1. In our baseline simulations, λ t ranges from 0.1 to 1.0 in increments of 0.1 as t goes from 1 to 10, simulating a gradual transition to a scenario where all new national PV capacity is domestically supplied. At the starting year of both the historical and projection scenarios, λ t = 0 and the cumulative capacity is set to the globally installed capacity in that year. Unit price declines under national market conditions thus evolve more slowly according to how rapidly λ t approaches 1. The national market scenarios propose that national-specific learning is proportionally derived from national versus global cumulative installed capacities, and by definition q it < q t . Extended Data Fig. 5 illustrates the relationship between λ it and the proportion of national to global cumulative installed capacity over all years for each country. It is noted that the same value of λ i does not translate to the same proportion of national learning for each country. For example, if λ i = 0.4, then the proportion of national learning is 15% in the United States, 44% in China and 40% in Germany.
Uncertainty in parameter estimates is propagated throughout all of our analyses using multivariate normal draws from the full covariance matrix of model parameters. Lower and upper bounds on results reflect a 95% confidence interval taken from the 2.5% and 97.5% percentiles from these draws.

Limitations
Learning-rate analyses, although widely used, are subject to critiques in terms of under-specifying learning mechanisms 40,44,45 . In our application of these models, we include exogenous factors that could influence module prices but are not directly linked to learning (for example, polysilicon prices). Otherwise, we estimate a single learning coefficient for each country that captures the average learning owing to a variety of nation-specific factors that contribute to learning, such as learning by doing (average plant size), learning by researching (research and development) and so on. Although other studies have estimated learning models that attempt to disaggregate learning into constituent components 46 , our research focuses on the nation-specific price implications of trade barriers. Data gaps and insufficient observations preclude explaining the contributing factors to learning in each country. This introduces potential biases if learning mechanisms are differentially affected by globalization. Given the concentration of PV panel manufacturing in China, it is possible that a portion of the learning in China was due to achieving higher economies of scale than manufacturers in the United States and Germany. If so, then the savings reported from the differences in the global versus national market scenarios may be overestimated, assuming that US and German manufacturers would have achieved similar economies of scale in a counterfactual scenario where national producers meet domestic demand. Three alternative models were estimated to disaggregate module production, installation capacity and average plant size. Those results are shown in Extended Data Tables 2-4. Improving ease of access to credit for solar projects, as reflected in declining trends in weighted-average cost of capital, has and will continue to have a large impact on reducing power purchase prices for solar 5,47 . Therefore, restrictions in capital flows following from nationalistic policies could lead to even larger costs on developers. Installed capacities are unchanged across scenarios, ignoring any effects of price elasticity of demand which, if incorporated, would result in fewer installations in higher-cost 'national markets' scenarios. Finally, the specific outcomes in terms of estimating savings from global versus national market scenarios are sensitive to simulation parameters, such as the number of years until all national capacity is domestically supplied. These parameters can be varied and the outcomes compared using an open-source application available at https://jhelvy.shinyapps.io/solar-learning-2021/.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
We compile a comprehensive dataset of historical solar capacity and component price globally and in the United States, China and Germany. All data are publicly available at https://doi.org/10.5281/zenodo.6989075. Global installed PV capacity and price data are from the open database of the International Renewable Energy Agency (IRENA) 18 . For the United States, solar capacity data are from the Solar Energy Industries Association (SEIA) 48 , and module prices are assembled from two sources: the Lawrence Berkeley National Laboratory (LBNL) 49 and the National Renewable Energy Laboratory (NREL) 16 . The LBNL data are used for the 2006-2018 period as this series ends in 2018, and the NREL data are used for 2019-2020 to extend the series to 2020. This was chosen because the NREL data only start in 2010, and thus the LBNL series covers a broader range (Extended Data Figs. 2-4). For China, both the installed capacity and module price data (2007-2018) were extracted from reports and presentations by the Energy Research Institute (ERI) 50 , and the 2019-2020 data were extracted from China Photovoltaic Industry Association where the historical data are identical to that of ERI 51 . For Germany, capacity data are from IRENA, and module price data were extracted from Fraunhofer ISE 52 . All prices are in 2020 US$, and we adopt inflation adjustments using the IMF (https://data.imf.org/) and exchange rates from the Federal Reserve Bank (https://www.federalreserve.gov/ releases/h10/hist/). Source data are provided with this paper.