Accounting is a wonderful tool for converting tautologies into useful information. Here, for example, is a tautology: when a company spends money, somebody receives that money. And here is a useful mental model that helps investors think about booms and busts, time industry cycles, and spot second- and third-order outcomes of news: one company's expenditures are, very often, another company's revenue. 

From a macroeconomic standpoint, this kind of modeling is a helpful way to look at what drives a country's growth. Put simply, some fraction of the total output of an economy will go to workers, and some fraction will go to people with capital. Often, the growth process entails increasing capital's share at the expense of labor. What this means can be described in two ways:

1. As the country gets richer, workers get better off, but not in proportion to how much the economy has grown. If GDP per capita rises 20% over some period, and wages are only up 10%, the average worker may have higher purchasing power, but half of the country's output has gone to capital rather than labor.

2. Higher returns to capital are a subsidy for reinvestment, and economies require a lot of reinvestment to keep going. Roads, railroads, ports, airports, power plants, factories, and homes are all long-lived assets with high upfront costs. For a country to have a lot of them, a smaller share of national income has to go to consumption so a larger share can go to investment instead. Importantly, shifting more returns to capital does not necessarily make all the capitalists rich (though it can have that effect!). It means there's a race to identify good investments fast since there's more money chasing them, and when this kind of policy continues for too long, the wave of capital looking for a return can end up subsidizing spending that simply doesn't make economic sense. For example, in China circa 1980, pretty much any piece of physical infrastructure was probably worth either fixing up or tearing down and rebuilding entirely, so the country got good returns from holding wages down while reinvesting the proceeds of exports. Now that China is a richer country, with lots of infrastructure, it's harder to find good homes for incremental money—but the money continues to flow.

But at a micro level, the way to use this model is to trace the impact of economic changes along the supply chain. If there's some company that directly touches consumer demand—a retailer selling goods, a hotel offering a place to stay, etc.—then fluctuations in that demand will quickly move up the supply chain to the producers required to meet that demand.

Higher demand for travel, for example, will eventually lead to demand for more planes and hotels, but in the short term the big impact is on labor. Think about it: hospitality is a labor-intensive industry, and those workers generally earn below-average wages, so when travel booms—when there is, for example, pent-up demand after a pandemic—one second-order effect is that the lowest-paid workers start making a bit more. And a second-order effect of that is that they can spend a bit more. High-income workers tend to save more money, and their savings rate goes up when they experience windfall gains. Lower-income workers are usually scrimping, deferring some purchases, and missing out on things they'd like to spend on, so higher wages for them tend to increase consumer spending.

A "supply chain" that connects consumers to labor, but ends there, is as simple as they come. In reality, most cases are a lot more like the kind of supply chain behind complex products like datacenters. As a preliminary note, it’s worth pointing out that median corporate profit margins are about 15.5%, so 84.5 cents of every dollar a company earns gets spent on something else, so when we talk about supply chain interactions, we’re talking about a big fraction of any given company’s impact on the world.

Datacenters are a complement to basically anything that involves moving information around or processing it, and that increasingly means they're a complement to just about everything. When there's more demand for digital services, or demand for better services, that higher demand means datacenter spending goes up.¹

But if you build a datacenter, you’re going to need chips, and chips are produced in fabs (exceptionally expensive manufacturing facilities, usually $15bn-$20bn at the leading edge, but Intel is planning a cluster of fabs whose total cost is $100bn). When fab utilization is low, new demand just means that existing fabs need to run extra shifts. But when utilization gets high enough, it means the world needs more fabs, and needs more $200m-apiece EUV lithography machines to fill them.

This tends to be the big takeaway from looking at the world from a supply-chain perspective. When there's slack in the system, or an ability to immediately respond to incremental spending, we see a pretty steady impact on every link in the supply chain: a surprise 1% increase in datacenter spending produces a 1% increase in spending on datacenter chips, which also leads the replacement of chipmaking equipment to tick up by about 1%—not because additional equipment was needed to increase supply, but because more is in use, which means more will need to be replaced. 

But when there isn’t slack in the system, a small incremental increase in final demand can produce massive changes in total production capacity. The rough way to approximate this is to look at the useful life of the relevant investment, invert it into a depreciation rate, and then compare changes in demand to that depreciation rate. So if there's some kind of asset that lasts for 10 years, another way to look at it is that in a given year, 10% of those assets are getting replaced as they wear out. A 2% increase in demand for whatever those assets produce, if they're all being used at full capacity, means a 20% increase in demand for the assets.

That rough approximation can work out in several ways. When there's an increase in underlying demand, the producers of whatever capital equipment ultimately bottleneck that demand can respond with some combination of:

1. Higher prices.

2. Longer lead times (just in case you were in the market for a passenger plane this year, best of luck with that—there's a shortage, and deliveries are delayed).

3. Additional capacity.

Industry cycles are partly driven by the fact that there's some game theory here: if every supplier choose option #1, customers complain but profits soar. However, if all of your competitors are choosing option #1, you naturally want to go with choice #3—you get to sell more, and prices are still high because you're not most of the market. Meanwhile, option #2 is an interesting choice, and also a cautious one. What it amounts to is telling favored customers that they'll get what they need, while other customers are out of luck.²

And this leads to some fascinating interactions. First, as lead times lengthen it encourages over-ordering: if you think you'll only get half of what you wanted on time, just buy twice as much. But over-ordering means that demand looks higher than it really is. Meanwhile, if the shortage lasts long enough, producers start to learn that they really can afford to produce more, so they do—supply expands! And at some point, supply expands fast enough to meet true demand, lead times start going down, double-orders start getting canceled (and some cash-conscious customers defer ordering entirely, knowing that they'll be able to buy on even more favorable terms if they wait). This is the classic "bullwhip effect" dynamic (Wikipedia article here, and there's a Diff piece talking about Covid and the Bullwhip Effect in 2020 as well).

Everyone acts on imperfect information all the time, and this leads every economic actor—households, investors, companies, governments—to frequently undershoot-until-they-suddenly-panic-and-overshoot, again and again. Thinking in terms of supply chains doesn't make anyone immune to this, but it's a good framework for asking whether the changes you see directly are a 1:1 measure of what's going on in the real world, or if the messy dynamics of variable lags and uncertainty are showing you a misleading view of the real world.

1. In fact, it doesn't have to be direct demand. There can be perceived demand: when Google decides to spend more money on datacenters, its competitors may suspect that they need to pull some of their own capital-intensive plans forward just to keep up, so one dollar of real incremental demand produces more than one dollar of incremental spending.

2. Why would an economically rational company do this? Because it's not their first cycle, and they know that when the cycle turns people will remember how they were treated.

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click that Get Started button below and we'll email you your link; if you are subscribed, be sure to check out the email version of this edition.

Help us out!

We'd like your feedback so we know what to cover, so if you haven't already, please fill out this one-minute form.