Experience curve
Passport
- Rock
- Mathematics
- Depth
- 3 · Drill rig
- Time to dig
- 3–10 years
- Capital
- ◉ · extreme
- Solo
- ✗ no
- AI
- ↑ AI-resistant
- Rent
- ✗ not for sale
Essence
Unit costs fall predictably — usually by 10–25% — with every doubling of cumulative output. The principle was formulated by Bruce Henderson and BCG in 1966, on the basis of the aviation learning curve (T. P. Wright, 1936: the fourth aircraft took 80% of the labour of the second). The difference from moat 15 is fundamental: scale economies are about current volume, spreading fixed costs, while the experience curve is about total output across all history and the knowledge that cannot be extracted from it. A competitor with the same volume today but nothing accumulated yesterday sits higher on the curve — and cannot catch up while the leader keeps moving.
How it is built
Semiconductors — yield as the sum of every wafer
Every wafer produced teaches: defects, calibrations, process windows. Where moat 19 is about that knowledge being unstealable, the experience curve is about it having a measurable price tag — yield and cost per transistor are functions of cumulative output. A late entrant is not merely less experienced; it is condemned to years of selling at a loss while passing through the same doublings.
Boeing against Airbus against COMAC
Thousands of airliners delivered make each next one cheaper to build: the duopoly has accumulated nearly a century of cumulative output. COMAC with the C919 has to walk the curve again — with state support, patience and unprofitable first hundreds of aircraft. Without state support, entering such an industry makes no economic sense at all.
SpaceX, and Wright's law in batteries
Launch cadence is the speed of descent along the curve: every launch teaches and makes the next one cheaper, and reusability turns a flight into a repeated lesson. A competitor flying twice a year descends physically more slowly even holding the same drawings. The same mathematics runs in Tesla and BYD batteries: the price per kilowatt-hour falls by Wright's law with cumulative production, and whoever has accumulated more is cheaper.
How it is bypassed
Change the curve instead of chasing it
A technological reset zeroes out what was accumulated: Chinese carmakers never chased a century-old combustion curve — they started on the electric one, where Toyota's and VW's experience in engines and transmissions barely transfers and where everyone started together on batteries. Every new S-curve is an amnesty: the leader of the old one begins the new one level with yesterday's newcomer, and carrying legacy on top.
Buy or siphon the experience
Experience is dissolved in people, suppliers and documentation: it is bought up through M&A and the hiring of dense groups — a bridge to process power — siphoned through equipment vendors, since Applied Materials and ASML distribute accumulated industry knowledge to all their customers and level the curves, and transferred through institutional mechanisms: Chinese joint ventures were for decades a conveyor moving Western carmakers' experience to local partners.
Move the learning into simulation
What used to be learned only on physical units is increasingly learned computationally: Waymo racks up billions of simulated miles against competitors' millions of real ones, and synthetic data and digital twins compress the cost of a repetition by orders of magnitude. The threat to the moat is asymmetric: where computation replaces physics, cumulative output stops being an advantage and the experience curve turns into a moat of compute budget — a bridge to capital.
Verdict
The quantitative brother of process power: knowledge bought with cumulative output, at a measurable price on the curve. Bypassed by changing the curve, by buying experience and by moving the learning into simulation — which is why, in the AI era, it migrates out of physics and into compute.