Competitive moats: Inside the factory vs. outside

In early America, nails were valuable enough to burn a house down for.

When a building had outlived its usefulness, people would sometimes burn it, sift through the ashes and recover the nails. For centuries, every nail had to be individually forged by hand. As late as 1810, nails accounted for an estimated 0.4% of US GDP.

Then came the machines.

Manual machines could produce around 10,000 nails a day. Steam power pushed that towards 30,000. Eventually, wire-nail machines produced hundreds of thousands.

Prices collapsed—spectacularly.

The lightbulb followed a similar path. A 60-watt bulb cost $1.75 in 1907—roughly $54 today. By 1930, Corning's ribbon machine could produce around 40,000 bulbs an hour. By 1942, a bulb cost just ten cents, while the cost of producing light had fallen by almost 99% from its 1882 level.

Both stories describe the same phenomenon: the learning curve.

In The Origins of Efficiency, Brian Potter explains how costs often decline as cumulative production rises—a relationship sometimes called Wright's Law. But the curve is not magic. Costs fall because workers learn, yields improve, processes are redesigned and scale justifies greater investment in automation.

In other words, the learning curve is accumulated knowledge.

Which raises an interesting investment question: when a company earns extraordinary returns, how much of that advantage comes from accumulated process knowledge—and how much comes from the environment in which it operates?

Income statements do not distinguish between the two.

Valuations sometimes do.

01The memory machine

Consider high-bandwidth memory, the specialized chips feeding AI accelerators. SK Hynix and Samsung sit at the centre of the current boom, and their advantage is not simply factory capacity. It lies in years of accumulated manufacturing knowledge—packaging techniques, yield optimisation and process know-how that cannot simply be purchased with a new machine.

Yet their valuations remain modest. Today, SK Hynix trades at roughly 7 times trailing earnings and Samsung at around 12 times.

The obvious explanation is cyclicality.

Memory prices rise when supply is tight and collapse when capacity catches up. But the industry's economics are also shaped by the learning curve: memory costs have historically fallen by roughly 28% with every doubling of cumulative production. Assuming long-term demand grows around 15% annually, each additional year of production pushes the industry further down its cost curve.

Normalised margins, therefore, are not static.

Using this framework, Samsung's valuation appears consistent with roughly five years of elevated profitability before margins fade; SK Hynix requires around four and a half. Management commentary, meanwhile, suggests the current supply shortage could extend until around 2028.

The market is therefore assuming a somewhat longer period of elevated profitability than management is promising—but not an absurdly longer one.

Now consider another industry that has travelled very far down its own learning curve: solar manufacturing.

China built enormous scale. Factories improved. Supply chains deepened. LONGi became the world's largest solar wafer producer for nine consecutive years.

And yet, it has lost money for the past two years—RMB8.6 billion and RMB6.4 billion in FY24 and FY25 respectively—while company-wide gross margins collapsed to just 0.81%.

One of the world's largest solar manufacturers is essentially selling at cost.

The problem was not inefficiency. China became exceptionally good at manufacturing solar products. But capacity expanded faster than demand, just as tariffs and other trade barriers in major import markets constrained access to overseas demand. The resulting surplus flooded the market. Prices collapsed. Competitors fought for volume.

The customer captured the benefit.

That is the paradox of efficiency. When everyone gets better at making the same product, the product becomes cheaper—and eventually, even the best manufacturers struggle to earn attractive returns.

02The protected factory

Now place this beside India's solar manufacturing industry.

Several listed Indian manufacturers currently generate returns on capital ranging from the high-20s to over 60%. At first glance, this looks like evidence of exceptional businesses.

But where do these returns come from?

India does not yet possess China's depth in wafers, ingots or polysilicon, and domestic manufacturers continue to import important inputs. Indian company managements openly acknowledge that imported cells can land in India at around $0.04–0.045 per watt, while domestic manufacturing costs are closer to $0.07–0.08. Domestic selling prices, meanwhile, are around $0.12–0.13.

The difference is not primarily inside the factory.

It is outside.

Approved manufacturer lists, customs duties and other trade restrictions have created a protected domestic market. Chinese manufacturers may produce cheaper products, but they cannot always sell them freely into India—or the US, for that matter.

The result is extraordinary pricing power for Indian manufacturers.

There is nothing inherently wrong with this. Virtually every country has, at some stage, used protection to give domestic industries time to develop capabilities and scale.

But investors should distinguish between a competitive advantage created inside the factory and one created outside it.

Both can generate high margins. Both can produce exceptional returns on capital.

But they are not the same asset.

For memory companies, a DCF can reasonably assume that some economic advantage survives because cumulative production continues to improve the cost base. Solar protection is different. If the policy framework changes, domestic manufacturers are exposed to the global cost curve—and China remains far ahead on that curve.

That makes the valuations particularly interesting.

For the Indian solar companies examined, the reverse DCF implies that today's elevated margins will barely fade, if at all. That may prove correct. Protection can last longer than expected, and companies can use that period to develop genuine capabilities—indeed, that may be the very purpose of industrial policy.

But the distinction still matters.

A factory can learn. Every doubling of production can improve yields, reduce costs and deepen accumulated knowledge.

Policy can create attractive economics, but it does not itself improve the underlying process.

Markets, of course, are natural extrapolators. Exceptional returns persist long enough and a temporary advantage can begin to resemble a permanent moat.

The danger is not protection itself.

It is paying for protection as though it were process.