Strategy frameworks
New technology starts expensive. The cost curve tells you how fast it gets cheap.
Lithium-ion battery cells cost 99% less than they did in 1991. What the cost curve is, why it happens, and how owners and investors should use it.
In 1991 a lithium-ion battery cell cost about $9,200 per kilowatt-hour of capacity, in today's dollars. In 2024 it cost $78. That is a 99% decline in 33 years, about 13.5% a year. New technology almost always starts at a premium. What happens next is more predictable than most business plans assume.
In brief
- New technologies get cheaper on a curve: each doubling of cumulative production cuts cost by a roughly steady percentage. The idea dates to aircraft plants in the 1930s.
- Lithium-ion cells and solar modules have followed the same curve for decades, each falling about 13% a year. Battery cells have fallen about 20% for every doubling of cumulative production, solar modules about 24% for every doubling of installed capacity.
- The curve is not smooth. Raw-material spikes interrupt it, as lithium did in 2022 and polysilicon did in the mid-2000s.
- For owners and investors, the curve is a forecasting tool, a pricing discipline and a warning: the savings are shared across an industry, so they are rarely a moat on their own.
The pattern has a name
In 1936 Theodore Wright, an engineer at Curtiss-Wright, showed that the labor needed to build an airplane fell by a steady share, about 20%, every time the total number built doubled. In the 1960s Bruce Henderson of the Boston Consulting Group extended the idea to all of a product's costs and called it the experience curve. Today it is often called Wright's law.
The mechanism is not mysterious. Each doubling brings more learning on the line, larger plants, better equipment, a deeper supplier base, design changes that remove cost, and occasional breakthroughs in materials or process. None of these happens by itself. Together, across an industry, they move cost down at a rate that has been surprisingly steady for many technologies.
The rate matters more than any single price. A 20% learning rate means cost falls by a fifth with every doubling of cumulative volume. Early on, when volume doubles every year or two, cost falls fast. Later, each doubling takes longer, and the decline slows even though the curve is unchanged.
Lithium-ion: 33 years on the curve
Sony began selling lithium-ion cells in 1991, for camcorders and other consumer electronics. Laptops and phones carried the volume through the 2000s. Electric vehicles and grid storage took over in the 2010s, and that is when volume, and the cost decline, accelerated again.
Researchers at MIT who rebuilt the price history from more than 1,700 records found that real cell prices fell about 13% a year from 1992 to 2016, and about 20% for each doubling of the cumulative market. Plotting price against the total amount of lithium-ion ever produced shows the curve directly:
Cumulative production went from about 130 kilowatt-hours in 1991, roughly two of today's electric-car battery packs, to about 3,500 gigawatt-hours by 2023. That is roughly 25 doublings. Across all of them, price fell about 20% per doubling. The shape matters too. In the 1990s, when total volume was tiny, prices moved slowly. Once production reached gigawatt-hours, the curve steepened to about 29% per doubling. Learning needs volume to work on.
The decline has continued since. BloombergNEF's 2025 survey put the average battery pack, the finished assembly that goes into a vehicle or a storage system, at $108 per kilowatt-hour. That is down 93% since 2010. Packs for stationary storage averaged $70.
The curve has also had a bump. In 2022, average pack prices rose 7%, the first increase since BloombergNEF began its survey in 2010, as lithium and other materials spiked. Prices resumed their fall the next year. Over a long horizon the trend won. In a single year, the commodity did.
Solar ran the same curve, for longer
Solar modules are the clearest long-run example. Batteries are priced per kilowatt-hour of storage and solar modules per watt of generating capacity, so they cannot share an axis. Plotted on the same timeline and the same log scale, where a straight line means a constant rate of decline, their slopes look almost the same:
A solar module cost about $132 per watt in 1975 and $0.12 in 2025, in today's dollars. That is about 1,100 times cheaper. Plotted against cumulative capacity installed worldwide, the price follows Wright's law almost exactly:
Over 22 doublings of installed capacity, the module price fell about 24% with each one. The points above the line in the mid-2000s were a polysilicon shortage. Prices stalled, then fell faster than the trend once new silicon capacity arrived.
It is not only energy
Computer storage ran the curve even harder. The cost of storing a terabyte on disk drives fell about 29% a year from 1956 to 2023, and memory fell about 35% a year. Henry Ford's Model T went from $850 in 1908 to under $300 by the mid-1920s, which we covered in How to forecast a market that doesn't exist yet. Hamilton Helmer notes in 7 Powers that most experience curves studied fall in a band where each doubling takes cost to 70 to 85% of its prior level. In other words, cost falls 15 to 30% per doubling.
What the curve means for owners and investors
1. Forecast the price, not just the volume. If your plan depends on a technology that is still on its curve, the price in year five will not be today's price. A fleet electrification plan, a plant energy-storage project or a new product that uses batteries should be modeled with a falling cost line. The same is true in reverse for a business case that assumes today's premium will last.
2. Price ahead of the curve, and fund the cost roadmap to meet it. Henry Ford set prices for where cost would be, not where it was, and let volume drive cost down to meet them. In the 1960s Robert Noyce did the same at Fairchild, pricing early integrated circuits below what they cost to make on the bet that volume would bring cost below price. A supplier selling a component on a technology curve faces the same choice. Its customers will expect annual price reductions. The supplier that plans its cost roadmap around the curve keeps its margin. The one that waits for the request gives it away.
3. Do not mistake the curve for a moat. Helmer's point is that experience-curve savings are largely shared. Equipment makers, suppliers and engineers carry the learning across the industry. A competitor that reaches the same cumulative volume gets similar costs. When a target's investment thesis rests on a cost lead from experience, test whether rivals with similar volume have similar costs. If they do, the lead is temporary. Lasting advantage comes from something the curve does not give everyone: scale in a business with high fixed costs, switching costs, or a process others cannot copy.
4. Model the bumps. Raw materials set a floor under cost, and commodity spikes interrupt the curve for a year or two. Build a scenario where a key input doubles, and decide in advance whether you would pass it through, absorb it or wait it out.
5. Expect demand to expand as cost falls. Each step down the curve brings in buyers who could not justify the purchase before. Battery packs near $100 per kilowatt-hour opened up electric vehicles and grid storage at a scale that was not possible at $1,000. A forecast that holds the market size fixed while cost falls will understate both.
How to build a cost-curve forecast
- Find the learning rate. Use the technology's own price and volume history if there is one. If not, use analogs. Most fall between 15% and 30% per doubling.
- Forecast cumulative volume. The curve runs on cumulative production, not time. A volume forecast, built from adoption analogs, turns into a cost forecast.
- Check it against a bottom-up cost model. Materials, labor, energy and capital per unit set a floor. As cost falls, materials become a larger share and the curve flattens toward that floor.
- Add the bumps. Run a scenario with an input-price spike and one with faster adoption.
- Update every year. Compare actual prices to the curve. A widening gap says something has changed in the technology, the supply chain or the market.
What to do on Monday
List the technologies your business makes, buys or competes against that are less than 20 years old. For each one, find the price five years ago and today, and work out the annual rate of decline. Then ask what your plan assumes for the next five years. If it assumes the price stays flat, you are probably wrong, and you should know in which direction that hurts you.
Monthly note
One short note a month on growth strategy, deals, and industrial markets. No spam, unsubscribe anytime.