The Learning Curve

I cut my teeth in the computer industry, where for decades Moore’s Law has pretty accurately described the way that year by year computers become more and more powerful for a given price. Or put another way, that every year transistors become cheaper and cheaper. Is this perhaps due to some magical property of transistors? No – as the picture below shows, it was happening before transistors were even invented, and it will surely continue to happen even after we abandon silicon for optical or quantum computing:

Figure 1 – Moore’s Law was operating even before transistors

It turns out that Moore’s Law is in fact just a special case of a much more general law of Engineering known as the Learning Curve (or the Experience Curve). This states that the more we build of something – anything – the cheaper it will get, as engineers constantly find new ways to make the production more efficient. In one pithy phrase, the Learning Curve neatly summarises the true power of Technology and of Engineering.

To see the Learning Curve we plot a graph of “cumulative unit production to date” against “cost-per-item today”. If we plot the vertical cost axis logarithmically, then the Learning Curve is a roughly straight line showing how the cost-per-item continues to drop geometrically year by year (i.e. the cost each year reduces by a certain percentage relative to that of the previous year, and that percentage tends to remain constant).

The concepts underpinning the Learning Curve were first described in the Journal of the Aeronautical Science in 1936 by T. P. Wright in an article about aircraft production. I believe the term Learning Curve was then coined in the 1960’s by Bruce Henderson of the Boston Consulting Group. Quite simply, the more aircraft that were built, the cheaper they got:

Figure 2 – the Learning Curve for aircraft production
Source: C. Lanier Benkard, American Economic Review 2000

It turns-out that the Learning Curve applies equally-well to pretty much everything else technological. To pick a random example, Japanese Beer Production: as the cumulative volume of beer bottles made grows, so the price continues to fall.

Figure 3: The Learning Curve for Japanese Beer Production, 1951-1968
Source: William D Eggers, Deloittes, after Wally Rhines

And it turns-out it’s true for steam turbine generators too. The more we made, the cheaper they became:

Figure 4: The Learning Curve for Steam Turbine Generators

The consequences

And therein lies a tale. In 1769 James Watt invented the world’s first efficient steam-engine, the precursor of those turbines. Now you might think that inventing a more efficient machine will result in a reduction in the input raw materials – after all, Watt’s steam engine could do the same work using less coal. But of course what it actually did was to light the blue touch paper on the industrial revolution. The machines were incredibly useful, so we built more of them. Thanks to the Learning Curve, building more of them made them cheaper and cheaper. So now more and more people could afford them, so even more were built, and so on.

The net result of all this is that world coal consumption has risen exponentially from that day to this. Shocking but true – yes, it’s still rising. And of course we now realise the consequences, as atmospheric CO2 rises in tandem:

Figure 5 – the rise of coal consumption, and consequent rise in atmospheric CO2
Source: Sustainable Energy – Without the Hot Air, p19


So the Learning Curve is a very powerful force. It has delivered all sorts of wonders to our everyday lives over the past 250 years, but it also has a dark side: it has wrought havoc with our environment.

So is there any way that we can use the power of the Learning Curve to dig ourselves out of this mess? Luckily, yes.

It turns out that the Learning Curve applies equally well to sustainable technologies too. Below we can see how the cost of renewable electricity from wind turbines is steadily falling as we build more of them:

Figure 6: The Learning Curve for wind turbines
Source: Bloomberg New Energy Finance, ExTool

That 14% number means that every time we double the cumulative production of turbines, the cost falls by 14%. Perhaps that doesn’t seem particularly significant in just one doubling, but like the grains of rice in the Biblical story it soon multiplies up. Over the course of the above graph, in less than 30 years we’ve gone from about 200MW to 200,000 MW total production. That is around 10 doublings in cumulative production, which means 10 lots of 14% reduction in cost, the cumulative effect of which is that costs have been driven down to roughly one quarter of what they were. And still they fall.

And the Learning Curve applies equally well to renewable electricity from Solar panels too. As with computing, the technologies may change but the relentless downward cost pressure continues:

Figure 7 : The Learning Curve in solar panel cost
Source: Ken Zweibel, GW Solar Institute, George Washington University

Hastening along the Learning Curve

Just this morning on the radio I heard Lord Lawson, formerly UK energy secretary, and then UK Chancellor in the 1980’s, and perhaps our most eminent climate change denier, railing about the “unaffordable cost of renewables”. Well yes of course renewables are expensive Nigel, but only because they haven’t yet had the benefit of fossil fuels’ 250 years’ Learning Curve. People concerned about the subsidies afforded to renewables need to understand that the purpose of these subsidies is NOT to build out a massive amount of renewables at today’s high costs. It is to hasten the Learning Curve along to the point where these technologies become cheap-enough to compete directly with their fossil-fuel cousins – so-called “parity” – at which point the market will take over and the learning curve will kick into a higher gear as they become the cheapest no-brainer choice.

The different between 20MW and 40MW of generation is insignificant to start with, but winding forwards a few decades it becomes for example the difference between 20% and 40% of the UK’s energy needs. Accelerating us through the early part of the learning curve now will pay a really big dividend later-on.


Ray Anderson was founder of Flor, a floor-tile company. A successful if perhaps not very sexy business you may think, but Ray was passionate about it and had an epiphany when he realised that his business was not, in the simplest most literal sense of the word, sustainable. It could not go on doing as it was doing, because its business was based on consuming our oil reserves, which are finite. So he set out to make it sustainable. He documented his journey not long before he died in an excellent TED talk which I’d highly recommend watching. In it he refers to the “equation” popularised by Paul and Anne Erlich in their 1960’s book The Population Bomb. I = P x A x T. Our Impact on the environment is the product of our Population multiplied by its Affluence multiplied by its Technology. So technology multiplies our destructive reach.

But as Ray describes so well (and I hope I’ve made a case for above), there’s absolutely no reason why technology T should be the bad guy. We now need to ensure that in everything we do, as founders, engineers, industrialists, technologists, scientists, policy makers and consumers, our technology moves to the denominator, reducing our footprint. And then the learning curve will do its steady magic.

Figure 8: “T2”
Source: TED talk by Ray Anderson

To us as individuals, a T2 Learning Curve looks just like a T1 Learning Curve. Both bring us more good stuff, cheaper. But to the planet T1 is the road to hell, and T2 is the way back again.

With special thanks to Keith Walker who originally drew my attention to the fact that Moore’s Law is just a special case of the Learning Curve.


2 thoughts on “The Learning Curve

  1. Pingback: More than Moore’s Law « Jason’s Blog

  2. Pingback: Moore's Law: How long will Moore's Law continue to hold? - Quora

Comments are closed.