Material Insights · Understanding the Material World

Cost, trade-offs and resources: why the best design is rarely the strongest: Why things get cheaper: learning, scale and standards

All courses / Understanding Engineering / Module 13: Cost, trade-offs and resources: why the best design is rarely the strongest Part 2 of 5

Why things get cheaper: learning, scale and standards

Solar modules, batteries and shipping containers have all become far cheaper per unit, but for different reasons. Some fell because makers learned by doing, some because plants grew larger, and some because everyone agreed to make the same thing. Telling the three apart is the key to guessing what will get cheaper next.

2.1 A learning curve cuts the price by a fixed share each time total output doubles

Every factory gets better with practice: fewer rejects, thinner wafers, shorter cycle times, cheaper suppliers. In the 1930s an aircraft engineer, T. P. Wright, noticed that the labour to build an airframe fell by a steady fraction each time the number built doubled. The same pattern, now called a learning curve or experience curve, appears across many manufactured goods. The learning rate is the percentage by which the price falls with each doubling of cumulative output.

Solar is the best-known case. The 17th edition (2026) of the International Technology Roadmap for Photovoltaics puts the learning rate for modules at 26 per cent for 1976 to 2025. A 26 per cent fall means each doubling multiplies the price by 0.74. Halving the price therefore takes ln 0.5 / ln 0.74 = 2.3 doublings (calculated). Batteries follow the same logic. BloombergNEF's survey found lithium-ion packs averaging US$108 per kWh in 2025, 8 per cent lower than in 2024. Packs for stationary storage averaged US$70 per kWh, the lowest of any segment for the first time, and packs in China US$84. Our World in Data finds that battery prices have fallen about 19 per cent with each doubling of cumulative production since 1998; taking a round 20 per cent, a halving needs about 3.1 doublings (calculated).

Two features of the arithmetic matter. First, doublings get slower. Global solar capacity was nearly 3 TW at the end of 2025 and about 0.7 TW was added that year, so one more doubling takes four to five years at today's pace, where an early doubling took months. Second, the curve is about cumulative experience, not calendar time, so the price is not guaranteed to fall in any given year. Module prices in fact rose by about one cent per watt in 2025, to near US$0.09, as makers in an oversupplied market tried to rebuild margins.

Data Worked estimate 4: what a learning rate implies for solar modules and battery packs
ItemSolar modulesBattery packs
Price in 2025about US$0.09 per W (end 2025)US$108 per kWh (all packs)
Learning rate assumed26 per cent (1976 to 2025)20 per cent (round figure; about 19 per cent since 1998)
Price multiplier per doubling0.740.80
Doublings to halve the priceln 0.5 / ln 0.74 = 2.3 (calculated)ln 0.5 / ln 0.80 = 3.1 (calculated)
After one doublingabout US$0.067 per W (calculated)about US$86 per kWh (calculated)
After two doublingsabout US$0.049 per W (calculated)about US$69 per kWh (calculated)
After three doublingsabout US$0.036 per W (calculated)about US$55 per kWh (calculated)

Time check for solar: cumulative capacity was nearly 3 TW at the end of 2025 and 706 GW of modules were shipped in 2025, so one more doubling needs about 3 / 0.7 = 4 to 5 years at that rate (calculated). These are projections from a rule, not forecasts.

Source: ITRPV 17th edition (2026); BloombergNEF (December 2025); Our World in Data (updated March 2026); IEA PVPS Snapshot 2026, learning rates, module and pack prices, cumulative solar capacity
  • ITRPV, International Technology Roadmap for Photovoltaics, 17th edition (2026), as reported by PV Tech, 25 June 2026 · learning rate 26 per cent for 1976 to 2025; module price about US$0.09 per W at end 2025; 706 GW shipped, https://www.pv-tech.org/itrpv-2026-solar-industry-maintains-historic-learning-curve-despite-market-turbulence/
  • BloombergNEF, Lithium-ion battery price survey (9 December 2025) · pack prices by chemistry, segment and region, 2025, https://about.bnef.com/insights/clean-transport/lithium-ion-battery-pack-prices-fall-to-108-per-kilowatt-hour-despite-rising-metal-prices-bloombergnef/
  • Ritchie and Rosado, Battery costs have declined by 99% in the last three decades, Our World in Data (2024, updated March 2026) · about 19 per cent price fall per doubling of cumulative production since 1998, https://ourworldindata.org/battery-price-decline
  • IEA PVPS, Snapshot of Global PV Markets 2026 · nearly 3 TW cumulative at end 2025, about 698 GW installed in 2025, https://iea-pvps.org/snapshot-reports/snapshot-2026/

2.2 Bigger plants cost less per unit because cost follows surface and output follows volume

Scale is a different source of cheapness from learning. Build a bigger version of the same thing once, and the cost per unit of output drops at once. The reason is geometry from the module on Thinking in estimates. A tank twice the volume does not need twice the steel, because its wall area grows more slowly than its volume. Piping, instruments, buildings and the people to run a plant also do not double when capacity doubles.

Estimators in the process industries turn this into the six-tenths rule: the cost of a plant is proportional to its capacity raised to a power near 0.6, with real exponents ranging from about 0.5 to 0.8 for different equipment. Doubling capacity multiplies the cost by 2 to the 0.6, which is 1.52, so cost per unit falls to 76 per cent. A tenfold plant costs about four times as much and delivers each unit at 40 per cent of the price. That is why refineries, steelworks and cargo ships have grown for decades.

Scale has limits and costs. A large plant ties up a great deal of money before it earns anything and takes longer to build, so interest on the construction period raises its cost (the effect of discounting from Part 1). And in some products size works against you: wind turbine blades grow longer, and the power captured rises with the square of rotor diameter while blade mass rises with its cube. Mass per kilowatt therefore grows with size, and engineers must find lighter materials or cleverer shapes to keep the gain (see the module on Structures and materials). Small modular products, such as solar cells, batteries and electrolysers, get cheaper by the number made. Large single plants get cheaper by size.

Data Worked estimate 5: the six-tenths rule, cost against capacity (cost proportional to capacity^0.6)
Capacity relative to the base plantTotal cost relative to base (calculated)Cost per unit of output relative to base (calculated)
1 (base)1.001.00
22^0.6 = 1.520.76
44^0.6 = 2.300.57
1010^0.6 = 3.980.40
100100^0.6 = 15.80.16

The exponent 0.6 is an empirical average and is only reliable over a limited range of sizes, often a factor of 10 either side. With 0.8 the saving from doubling is only 13 per cent.

Source: Six-tenths rule: standard process-industry estimating rule, after R. Williams, Chemical Engineering (1947), cost proportional to capacity^0.6; ratios calculated here
  • Kosky et al., Exploring Engineering (Academic Press, 2013) · ch. 5, 'Engineering Economics' (finance charges during a long construction period)

2.3 Standards spread one design's cost across many users and let parts from different makers fit

The third source of cheapness is agreement. A standard fixes dimensions, grades, test methods or interfaces so that parts made in different places can work together. The shipping container is the clearest example. Its sizes, 20 and 40 feet long and the same width, and its strong corner fittings, are fixed by international standards, so the same box can be lifted by any crane, stacked on any ship, carried on a train wagon or a lorry and sent through any port. The box is a poor shape for some cargoes, but it carries almost everything at a low and predictable cost. Look it up under Shipping container and Cargo ship.

Standards also make learning curves possible, because many buyers and makers feed experience into one design. A household in India can fit an LED lamp in an old bulb holder because the cap fittings are standard. Reinforcing steel is sold in graded strengths under Indian Standard 1786, so a site engineer knows what Fe 500 bar will do before testing it. A standard also stores knowledge: codes record what past failures and tests taught (see the module on Design, failure and safety).

The economics of a standard run through fixed costs. A product with a tooling cost spread over more units has a lower unit cost, which is the same arithmetic as scale. The price is lock-in. A standard size is almost never exactly the optimum for a given job, and a standard that is wrong can persist for decades.

Value chain Three routes to a lower unit cost, and what each gives up
  • Lower cost per unit of output
    • Learning (experience) price falls with cumulative output; needs a big, competitive market; gives up nothing but can stall at material costs
    • Scale (size) cost grows more slowly than capacity; gives up flexibility, adds financing risk and a square-cube limit
    • Standardisation (agreement) one design shared by many users; gives up the perfect fit and risks lock-in
Source: Summary of Part 2 of this module, learning, scale and standardisation as described in the three sections
  • Kosky et al., Exploring Engineering (Academic Press, 2013) · ch. 5, 'Engineering Economics' (fixed cost recovered over units sold: the break-even point)
  • Hammack, The Things We Make (Sourcebooks, 2023) · ch. 2, 'Best: The Vitruvian Man, Kodachrome, and the Constant Search for Better'

See it in the table

St SteelAg SilverLi LithiumElz Electrolysers

Check your understanding

A module price follows a 26 per cent learning rate. After one more doubling of cumulative output a US$0.09 per W module should cost about:

A 26 per cent fall means multiplying by 0.74, giving 0.09 x 0.74 = 0.067. Halving would need a 50 per cent fall, which takes about 2.3 doublings.

Why does the six-tenths rule make big plants cheaper per unit?

With an exponent of 0.6, doubling capacity raises cost by only 52 per cent, so each unit of output costs 24 per cent less. The interest on a larger construction period works against this, so it is a tendency, not a guarantee.

Which of these gives up something for lower cost by agreement, not by size or by experience?

A standard trades a perfect fit for interchangeability and volume. The loss is usually small, but lock-in to a poor standard can last a long time.
← Part 1: Whole-life cost and the price of moneyMark read · Part 3: Choosing a material under cost, weight and carbon →