Economies of scale

A company that gains a scale advantage has to decide what to do with it, and the two obvious choices lead in different directions. Ford passed savings to customers, cut the price, and used the resulting demand to run the plant harder. Netflix held revenue per member near $11.70 a month and kept the difference: operating margin rose from 18% in 2022 to 27% in 2024. Price cuts widen the gap over rivals and are hard to reverse. Retained margin funds the next investment, but it leaves room for a competitor to undercut.

The number that governs the decision is utilization, not size. TSMC says so in its annual filing to the SEC. Because it owns most of its factories, a significant portion of its operating costs is fixed. When utilization rises, those costs are spread over more output and margins improve, and when demand falls, margins can drop significantly. Before committing to a large fixed cost, a leader should be able to state the volume that fills it, how soon demand can arrive, and what happens if it arrives late.

Average cost also conceals the cost of the next unit. Once Highland Park was built, one more Model T was cheap to make, so the plant’s average kept falling as output climbed. When demand outgrows a facility, the next increment needs another one, and average cost can jump. The useful comparison is the return on the next block of capacity, measured against demand you can see, and not the average across assets already built.

The scale that made the Model T cheap also committed Ford to a single design, and the design outlived its market. By January 20, 1926, Ford’s vice president Ernest Kanzler was urging Henry Ford in a memo to end the Model T because sales were falling, and Ford answered by pushing Kanzler out within months. Chevrolet passed half a million cars in 1926. In May 1927, after about 15 million Model Ts, Ford shut down for roughly six months to build the Model A. According to PBS, almost 75% of existing tools had to be scrapped, rebuilt or refurbished, and 60,000 workers were laid off. Chevrolet outsold Ford by more than two to one that year, and Ford never again dominated the low-priced field.

The lesson is not that Ford scaled too far. Its cost curve was tied to a product that could not change without rebuilding the plant. Chevrolet did not need to match Ford’s cost per car to take its customers; it needed a car buyers wanted more. Volume lowers the cost of making more of the same thing, and it raises the cost of making something different.

Fixed costs also work in reverse. TSMC’s own filing warns that its fixed costs do not decline when demand or utilization drops, so margins can fall sharply. Netflix shows the tighter version: payments for content assets rose 29% in 2024, from $13.1 billion to $17.0 billion. The scale advantage held because revenue grew 16%. If membership growth or pricing stalls while content spending keeps climbing, the same fixed cost that lowered the average starts raising it.