Learning Curves

Over the last ten years, the surface of the Earth warmed by another 0.5°C. At the same time, renewable energy grew its share of world electricity production from 5 to over 11 percent. These are the basic coordinates for the ongoing turn in global climate policy—from meting out taxes and penalties to promoting technologies that can substitute fossil fuels. Dealing with climate change feels more urgent than ever and also—for the first time since it emerged as an object of global politics in the 1990s—like a technical and economic possibility.

As a result, the boundaries between climate policy and economic policy are more porous than ever. Countries around the world are recasting their strategies for claiming a greater share of global value-added in the language of the energy transition, and launching new subsidy and incentive programs to support domestic industries with green credentials—the USA’s Inflation Reduction Act is the paradigmatic example. It’s no surprise that organizations ranging from ExxonMobil and OPEC to Bloomberg New Energy Finance (BNEF) and the International Renewable Energy Agency (IRENA) produce so many forecasts, roadmaps, and scenarios predicting how much energy the world will consume in 2050, and where that energy will come from.

The headline takeaways from these modeling exercises vary widely, with some predictable differences—ExxonMobil predicts that demand for oil and gas will grow 13 percent by mid-century, while BNEF predicts it will fall by 7 percent. However, they share a family resemblance. All of them rely on some kind of least-cost optimization framework, solving for the cheapest, most efficient global energy system possible, given whatever external constraints modelers choose to impose, like a maximum level of emissions. While modelers represent the global energy system at different levels of complexity, from one system planner with perfect foresight to multiple agents with distinct preferences, all of these approaches require some notion of costs to decide which technologies should be used to meet energy demand and which should be left on the shelf.

The most pervasive way of representing how costs will change over time is through the concept of the experience curve—an idea that followed a strange, circuitous path through the twentieth century on its way to become a load-bearing element of thinking on climate policy. The “experience curve” (sometimes called the “learning curve” or “Wright’s Law”) concept was first codified by the aircraft engineer T.P. Wright in a 1936 paper titled, modestly, “Factors Affecting the Cost of Airplanes.”

Wright drew the experience curve as a straight line on log-log paper, with axes representing the labor cost of producing a certain model of plane and the cumulative number of planes produced. In short, he was making the empirical observation that certain kinds of costs declined at the same rate each time production doubled, a ratio since dubbed the “learning rate” of a technology.

Over the ensuing decades, other authors found that experience curves seemed to fit the cost trajectories of technologies as varied as DRAM, PVC, and magnesium. Auspiciously, the concept was also championed by the Boston Consulting Group (BCG), starting in the 1960s. BCG used the experience curve in a new setting, taking a tool originally designed for cost accounting within a single firm and making it a bigger-picture framework for competitive analysis. It also expanded the definition of costs to include “R&D, sales expense, advertising, overhead, and everything else.” Despite these changes in scope, the concept came with a seductive rule of thumb that dates back to Wright’s original paper—that learning rates cluster around 20 percent.

Experience curves have long been controversial among economists, who tend to argue that they conflate learning-by-doing with factors as varied as economies of scale and a firm’s bargaining power with suppliers, and that they often represent a spurious correlation rather than a causal relationship. Nonetheless, they continue to be in wide use by industry and policymakers. By the late 1990s, the International Energy Agency (IEA) was promoting the use of the experience curve as a tool for policymakers sizing “learning investments” in new technology.

Early examples of government policy to drive clean-energy adoption—solar heating for swimming pools in Germany, distributed solar panel installations in Japan, and wind turbines in Northern Europe—showed that cleverly designed subsidies could help new technologies realize self-reinforcing cost reductions, eventually finding “docking points” in niche markets and, from there, reaching full commercial scale. Twenty years on, the language of “learning investments” and “docking points” has transformed into calls for “catalytic capital” to drive “commercial lift-off” but the thinking behind it is largely the same.

The experience curve concept really does map well onto reality for certain technologies. Over the last decade, the cost of utility-scale solar projects fell by 70 percent, and the cost of onshore wind farms by 39 percent. Lithium-ion batteries have fallen in cost by 82 percent. Solar, wind, and battery manufacturing are all likely to keep growing apace—BNEF estimates that these three technologies will see installed capacity grow by 4.6 TW by 2030, as compared to 2.1 TW of total capacity in-place today. At the same time, green industrial policy is increasingly directed at a range of earlier-stage technologies—“green” hydrogen, carbon capture and storage (CCS), long-duration energy storage (LDES), and advanced nuclear—where future learning rates are much less certain.

Not all technologies “learn” much with scale. Construction costs for nuclear power plants rose alongside cumulative capacity in many countries from the 1970s to early 1990s, including the US, Japan, and West Germany, while more recent nuclear build-outs in India and South Korea have been only a little more successful, managing to hold costs flat or slightly down (by 1 to 2 percent per annum) in real terms. We need to be clear-eyed about whether support for a specific industry is a down payment on future cost reduction, or a carbon tax by other means.

The ongoing crisis facing the offshore wind (OSW) is a case in point. Year-to-date, developers have canceled over 7 GW of planned OSW generation capacity, with more projects delayed or “under review.” For a sense of scale, only 8.4 GW of OSW was installed in all of 2022. Though projects in other countries are getting cancelled too, the OSW crisis centers on the US, where the Biden administration hopes to grow the industry from a standing start, adding 30 GW of capacity by 2030.

The root cause of the industry’s problems is that OSW projects take a very long time to reach fruition. In the US, the Sunrise Wind project, a joint venture between Ørsted and Eversource Energy, initially leased an area fifteen miles off Rhode Island from the Department of Energy’s Bureau of Ocean Energy Management (BOEM) back in 2013, won a deal to sell renewable energy credits to New York State in the summer of 2019 (which it will potentially re-bid), and offshore construction is slated to begin in April 2025. That implies an over six-year journey from pricing the clean-energy credits sold by the project to the start of offshore construction work, let alone “first power.”

Implicitly, this was a highly leveraged financial bet on the spread between inflation and the OSW industry’s learning rate. In 2019, when Sunrise Wind’s energy credits were priced, the unsubsidized cost of OSW power was about $150.¹ If a project locked in revenue at that price point, with no recourse to inflation adjustments, and construction costs ended up just 10 percent higher than planned, it would cut the rate of return for equity investors in half. Developers do try to limit their exposure to unexpected cost increases by locking in prices with key suppliers, like turbine manufacturers and construction vessel owners, well in advance, and, in certain cases, hedging their exposure to changes in the cost of key raw materials like steel, neodymium and copper. But, in 2019, before the post-pandemic commodity squeeze, the cost of wind turbines had fallen by nearly 40 percent over the last six years. It would only be rational to retain some of the risk related to changing prices for turbines and other inputs if you extrapolated that trend into the future.

As we now know, this was the wrong bet. Procurement processes for OSW off-take agreements are adjusting to this new reality—for example, the latest New York state OSW solicitation added a one-time inflation adjustment feature to the awarded contracts. OSW project developers like Ørsted have had to write off billions of dollars of assets after walking away from projects that no longer make financial sense. And the political opponents of OSW smell blood in the water.

Whether that is a “crisis” has little to do with what has actually happened with the cost of OSW power, and more to do with the expectation that costs would rapidly fall. The reality is that if we look at the most mature market for OSW (Europe), construction costs for OSW installations have fallen by a little under ~$300 per kW a year, in real terms, since 2010, but there has not been a pronounced trend in either direction since about 2018, just a lot of ups and downs. Interest rates have gone up, and a number of key commodity prices, like steel, are well off their highs but still elevated versus pre-pandemic levels. Putting all this together implies an unsubsidized cost of offshore wind somewhere in the $120-150 range, still higher than typical peak power prices in the US or corporate power purchase agreement (PPA) prices.

So this is a maturing industry, that may never be competitive with new-build gas-fired power plants on an apples-to-apples, unsubsidized basis. That doesn’t mean that it’s not deserving of government support. The Northeast US has few decent options for low-carbon power, with less sunshine than the Southwest and slower wind speeds than the Great Plains. New nuclear power plants face political opposition, and recent US experience with nuclear projects (i.e. the Vogtle project in Georgia) has been disastrous, with fully-loaded construction costs exceeding $10,000 per kW, over three times the cost of new-build nuclear reactors in Asia. Calculating the “social cost of carbon” is more art than science, and using the more aggressive estimates recently floated by the EPA would suggest that power from modern natural gas plants is underpriced by some $70 per MWh. Significantly, this implies cost parity with OSW, if CO₂ emissions were to be fully taxed at this price.

Nonetheless, this is not how OSW, and, indeed, green industrial policy as a whole have been sold to the public—as a strategy for incubating domestic low-carbon industries that will soon be able to stand on their own two feet. The key point is that we have to put the experience curve in its proper context, as just one intellectual tool, with its own particular history and baggage. Otherwise we risk getting sideswiped by cases where it doesn’t apply.

Not every industry we need to scale to decarbonize our economy will have a high enough learning rate to reach cost competitiveness with incumbent, fossil-fuel-based technology. And in some cases, the experience curve concept may not apply at all, given it is better suited for thinking about routinized manufacturing processes than site-specific mega-projects. Green industrial policy is well suited to saving promising, but expensive technologies from being “locked out”—a real case of innovation market failure. Over the next decade, we are going to learn in real time which technologies it works well for, and which it doesn’t. Building a political coalition that can see this new, more aggressive kind of climate policy through to 2050 will require accepting that a risk-taking, market-shaping state is not going to win every bet.