In a recent tweet (Aug 14th, 2018, reproduced below), Bill Gates said that supply demand curves don't work for software because the first unit costs a lot and every unit after that is free. Isn't this just modeled by a horizontal supply line?

At the time I was in college, this was basically how the global economy worked...

enter image description here

This chart assumes that the total cost of production increases as supply increases. When a car company makes a new vehicle, the 10th car costs the same to manufacture as the 1000th.

But software doesn’t work like this. Microsoft might spend a lot of money to develop the first unit of a new program, but every unit after that is virtually free to produce.

Unlike the goods that powered our economy in the past, software is an intangible asset -- “something you can’t touch.” And it’s not the only example: data, insurance, e-books, even movies work in similar ways.

The portion of the global economy that doesn’t fit the old model just keeps getting larger. But the rules that govern the economy haven’t kept up. This is one of the biggest trends in the global economy that isn’t getting enough attention.

If you want to understand why this matters, I highly recommend the new book “Capitalism Without Capital” by @haskelecon and @stianwestlake. Here’s my review: link.


5 Answers 5


Is Bill Gates wrong about supply demand curves?

Yes, this time he is wrong (and please know that, generally speaking, I identify myself as an admirer of Bill Gates). It is unclear to me how much of this comes from his own reflection versus from the book he reviewed.

First, supply [curve] does not depend only on the cost of production. It may also depend on other factors such as sake of profit and market-share strategy. The nature of "intangible" products might certainly involve different dynamics and different prevalence of these factors, whence conceptualizing supply exclusively on the basis of "how much it costs to produce an additional unit" is an oversimplification.

Second, Bill Gates's review appears to trivialize the concept of production costs, as costs do not cease once a product is released to the market. Henceforth, the company or supplier needs to allocate resources for maintenance & support. Larger numbers of units sold make it likelier that more vulnerabilities will be discovered, which in turn requires the hiring of sufficient staff so as to avoid bottlenecks and a subsequent loss of market share. In the software industry, this is (or was) palpable --for instance-- in the release of patches & updates to mitigate buffer overflow exploits.

Similarly, gaining a larger market-share subjects the supplier to both stronger scrutiny by authorities and a higher likelihood of expensive litigation of various sorts: unfair trade practices, monopoly, product liability, to name some. This and the preceding paragraph occur both in the context of both tangible products and "intangible" ones such as a new version of Windows OS.

Although it is very unlikely that maintenance and litigation costs will be proportional to the number of units sold (e.g. produced), the quantitative significance of these aspects are most recently evidenced by the falling of Facebook in the stock market since the wake of the Cambridge Analytica scandal.

It is hard and maybe futile to outline with precision the dynamics of supply of intangible goods. However, for the reasons explained above, an horizontal supply curve definitely would not reflect or capture the complexity of the market of intangible goods.

  • 2
    $\begingroup$ I'm a bit confused how your last paragraph ties into the rest of your answer. You begin by (quite rightly) pointing out that marginal costs are non-zero, even in the case of software (and similar goods). Why do you then conclude that "a horizontal supply curve" would be improper? A horizontal supply curve just means constant marginal costs- not non-zero marginal costs. The story you describe in the first 5 paragraphs points toward a case of positive, but certainly diminishing, marginal costs. That, in turn, would suggest a downward sloping LRS curve, not an increasing one. $\endgroup$
    – AndrewC
    Aug 24, 2018 at 20:40
  • $\begingroup$ @AndrewC Thanks for your comment. As for the positive slope of supply curve, I attribute it to (1) investors' sake of profit; (2) expectations that many consumers would be price-takers rather than fall behind in the market of that intangible good; and (3) a [big] company's greater exposure to liabilities (such as expensive litigation that won't necessarily preclude adverse judgments) insofar as there is much more scrutiny on Google, Microsoft,etc. than on competitors with a significantly smaller market share. In light of these aspects, I would be very surprised if supply curve were horizontal. $\endgroup$ Aug 24, 2018 at 21:32

No, Bill Gates is correct here. The simple econ 101 cross model is based on a separation of the supply and demand side. Firms are price takers, they assume they can sell as much as they want at the prevailing price and their output decisions have no impact on the price. If firms are not price takers, they have to explicitly take into account how the demand side reacts since this determines the impact on prices. Now with huge returns to scale, price taking is usually not a sensible assumption and therefore the econ 101 cross an inappropriate model.

  • 1
    $\begingroup$ I don't know about yours, but my econ 101 definitely mentioned economies of scale (and monopolies). for the former see e.g. Mankiw's textbook, 4th ed., p.281; monopoly on p. 311. Maybe at some community college these topics get skipped. $\endgroup$ Aug 26, 2018 at 4:16
  • $\begingroup$ Quote from the latter page "Microsoft has no close competitors and, therefore, can influence the market price of its product. While a competitive firm is a price taker, a monopoly firm is a price maker." What probably does not get explained well enough in such introductory classes is the link between economies of scale and the emergence of monopolies. There are some books which treat them closely though, e.g. open.lib.umn.edu/principleseconomics/chapter/… has a "Figure 10.1 Economies of Scale Lead to Natural Monopoly". $\endgroup$ Aug 26, 2018 at 4:24
  • $\begingroup$ Where in my answer did I write that undergraduate textbooks do not mention returns to scale and monopolies? $\endgroup$ Aug 26, 2018 at 7:28
  • 1
    $\begingroup$ I have answered the question "Is Bill Gates wrong about supply demand curves?" The question was not "Is Bill Gates wrong about the history of the world economy?" I noticed that your answer contains a diagram, but not one involving supply curves. $\endgroup$ Aug 26, 2018 at 7:53
  • 2
    $\begingroup$ This is the correct answer and it is particularly frustrating to see it so low on the page when the question (and answer) is such a basic one. $\endgroup$
    – Ubiquitous
    Dec 5, 2018 at 12:24

Gates posted in his Tweet some linear supply and demand curves (quantity vs. price) and then said

This chart assumes that the total cost of production increases as supply increases. When a car company makes a new vehicle, the 10th car costs the same to manufacture as the 1000th. But software doesn’t work like this. Microsoft might spend a lot of money to develop the first unit of a new program, but every unit after that is virtually free to produce.

Gates seems rather ignorant that (internal) economies of scale exist in many other industries. E.g., textbook-reminder material from The Economist:

Economies of scale are factors that cause the average cost of producing something to fall as the volume of its output increases. Hence it might cost \$3,000 to produce 100 copies of a magazine but only \$4,000 to produce 1,000 copies. The average cost in this case has fallen from \$30 to \$4 a copy because the main elements of cost in producing a magazine (editorial and design) are unrelated to the number of magazines produced.

Economies of scale were the main drivers of corporate gigantism in the 20th century. They were fundamental to Henry Ford's revolutionary assembly line, and they continue to be the spur to many mergers and acquisitions today. [...]

Internal economies of scale arise in a number of areas. For example, it is easier for large firms to carry the overheads of sophisticated research and development (R&D). In the pharmaceuticals industry R&D is crucial. Yet the cost of discovering the next blockbuster drug is enormous and increasing. Several of the mergers between pharmaceuticals companies in recent years have been driven by the companies' desire to spread their R&D expenditure across a greater volume of sales.

The software industry may be just an extreme example, where R&D costs are much higher than other costs, but the other costs are certainly not zero.

If a firm offers any customer support whatsoever for their software, it can't be the case that the support cost is completely independent from the number of units sold. Even if it's just scaling support websites so they take millions of page hits from self-help customer, it's still a cost. However, leaving aside his apparent poor knowledge of other industries, Gate's argument is that per-units costs are negligible ("virtually free") in comparison with R&D, not zero.

It's also not clear that sales and marketing costs don't factor substantially in unit costs for software. After a firm obtains a monopolistic position so it can afford to not bother with any marketing (although that's probably not the case even for Microsoft), maybe those costs do become negligible. Hower, as far as the average softare firm goes, a yahoo survey found that:

On average, [software] companies spend 15%–25% of their revenues in sales and marketing activities. However, there are exceptions like Symantec (SYMC) which has consistently spent ~40% of its revenues in sales and marketing.

So at least in some niches it seems that marketing costs are substantial. I don't expect these marketing costs to be independent of the number of units sold (and thus "produced").

And also from the same source on R&D costs:

R&D costs typically form 10%–20% of the revenues for software companies. However, not all of this goes into innovation. A large portion of it’s spent in testing various configurations of operating systems (or OS) instead of developing new functionality. Industry experts think that even less than 5% of R&D budget is spent on innovation.

Gates is also ignoring a more subtle issue: the R&D cost itself has some relationship to the market share of the software thus developed. Why? You cannot attain a monopolistic position with trivial software that's quick to write and thus offers a low barrier of entry to competition. So the size of the software is related to its ultimate market share (which translates into units sold), and so R&D costs are in this way a determinant of the number of units sold (so "produced"). The exact shape of this relationship is probably more difficult to quantify; empirical evidence has pointed to both economies and diseconomies of scale relating software size to its production cost. (Diseconomies are supposedly of the communication-complexity kind between [more] developers.)

Diseconomies of scale are not unique to software development either. The aforementioned article in The Economist talks about it a general context of organizations:

The larger an organisation becomes in order to reap economies of scale, the more complex it has to be to manage and run such scale. This complexity incurs a cost, and eventually this cost may come to outweigh the savings gained from greater scale. In other words, economies of scale cannot be gleaned for ever.

A related issue that's worth mentioning: Gates probably thinks his software support infrastructure is "virtually free" of cost because he already has it in place for new products. And that brings us to economies of scope. As nicely said (again) in The Economist:

First cousins to economies of scale are economies of scope, factors that make it cheaper to produce a range of products together than to produce each one of them on its own. Such economies can come from businesses sharing centralised functions, such as finance or marketing. Or they can come from interrelationships elsewhere in the business process, such as cross-selling one product alongside another, or using the outputs of one business as the inputs of another.

Gates' tweet doesn't talk about demand, but he does mention it in his extended notes he links to. These notes focus on the so-called "intangible investment" in R&D but also in marketing etc.

And when talking about those, he doesn't sound so radical after all. The pricing of products developed through heavy R&D costs (in particular medication) is a huge issue, with lots of research done on how to recoup the sunk R&D cost, or how to finance future R&D, e.g. by hoarding cash from current sales (as done by firms like Apple, Google, etc.) or other means (e.g. venture capitalism).

Gates is also fairly correct when he criticizes the "magical" belief in equilibrium as the best thing for society, laid up a as an easy target in the following terms:

Equilibrium is magical, because it maximizes value to society. Goods are affordable, plentiful, and profitable. Everyone wins. [Critique based on the software industry follows.]

And in this respect Gates is correct again; equilibrium is affected by economies of scale, although Gates omits (even from his extended notes) the typical/textbook effects that [internal] economies of scale have on equilibrium:

Internal ES [economies of scale]:
– Problem: it forces us to move to a imperfect competition model:
• In imperfect competition, firms can affect the price.
– Industries with few producers
– Industries with product differentiation for consumers
– In order to sell more, the price must be decreased (strategic behaviors)
• Possible cases:
– Monopoly; oligopoly; Monopolistic competition

So as a sort of conclusion: Gates gives some rather bad examples from other industries (e.g. saying that the auto industry lacks economies of scale is laughable), and he fails to talk about the elephant in the room (monopolies)... but his conclusion that equilibrium isn't necessarily a good thing from some points of view (particularly in the case of imperfect competition, engendered by economies of scale)... is actually correct, even though Gates' own argumentation for this is rather unconvincing. Keep in mind that his main point with tweets/blog is to promote a book and to argue that the "intangible investment" lacks sufficient recognition. (His last point reminds me of the controversy --which also involved Gates at one point -- whether there are enough developers; it depends who you ask.)

And let's dispel this myth that economies of scale or monopoly's effect on the [lack of] equilibrium are not covered in econ 101. E.g. a freely available textbook has a "Figure 10.1 Economies of Scale Lead to Natural Monopoly".

enter image description here

It also talks of sunk costs as a barrier to entry after that.

And Mankiw's textbook has had an "FYI: Why a Monopoly Does Not Have a Supply Curve" for many editions:

You may have noticed that we have analyzed the price in a monopoly market using the market demand curve and the firm’s cost curves. We have not made any mention of the market supply curve. By contrast, when we analyzed prices in competitive markets beginning in Chapter 4, the two most important words were always supply and demand. What happened to the supply curve? Although monopoly firms make decisions about what quantity to supply, a monopoly does not have a supply curve. A supply curve tells us the quantity that firms choose to supply at any given price. This concept makes sense when we are analyzing competitive firms, which are price takers. But a monopoly firm is a price maker, not a price taker. It is not meaningful to ask what amount such a firm would produce at any given price because it cannot take the price as given. Instead, when the firm chooses the quantity to supply, that decision (along with the demand curve) determines the price. Indeed, the monopolist’s decision about how much to supply is impossible to separate from the demand curve it faces. The shape of the demand curve determines the shape of the marginal-revenue curve, which in turn determines the monopolist’s profit-maximizing quantity. In a competitive market, each firm’s supply decisions can be analyzed without knowing the demand curve, but that is not true in a monopoly market. Therefore, we never talk about a monopoly’s supply curve.

Both are introductory "econ 101" textbooks.


Yes, Gates is wrong

If there was just one customer for a piece of software, the software would be cheaper than for two customers. With just one customer, it can work in just one way. For the second customer, it has to be more flexible.

You can see some quick examples of this in Microsoft Windows 10. It has seven different versions for different customers. Beyond that, it has a multitude of configuration options. Yet a single customer will tend to use just one configuration and stick with that.

Gates is correct to say that software is different. It does have greater returns to scale than cars. But cars also reduce price with scaling and face the problem of needing to be configured for multiple uses. Software isn't unique. Books always had a similar problem (writing the book is a huge part of the initial cost while actually printing the book is much cheaper). E-books accentuate that, but it's still the same problem.


Gates is laughably wrong about insurance. Insurance is not all first customer cost. Insurance improves with scale more like cars do. The first customer is exceptionally risky, as the company either has to pay out or not. If not, the insurance premium is all margin. If the company does pay out, then it almost certainly loses money on that customer. Additional customers amortize that risk. The more customers, the lower the risk, even though the payout increases.

Insurance companies work because customers can pool their risk together. Customers pay a small amount of money so that if an unlikely event happens, they can avoid paying out a large amount of money. The larger the pool, the lower the risk. But even a gigantic pool still adds payout, on average, with each customer. The payout becomes more predictable, not less in amount.

If there is a one in a hundred chance of something happening that costs \$1000 and an insurance company has one customer, then the premium needs to be really high. There's an argument that it needs to be \$1000, as otherwise the company won't have enough money to pay off the claim if it occurs. But if the company has a hundred thousand customers that cost \$5 to obtain and maintain, the company could charge \$15 per customer. It would expect to pay \$1 million on that many customers, or \$10 per customer. Add the \$5 and there you go. The company collects \$1.5 million and pays out \$1 million. The remaining \$.5 million goes to staffing and other overhead.

That's a simplified model, but it's how supply works in insurance. The first customer is the most expensive because the first customer is also the riskiest. After a while, the marginal cost settles to an approximately constant value.

Cars work the same way. Initially there are design and prototyping costs. The initial car may be made by hand rather than assembly line. So the first car, the concept car, is really expensive. Subsequent cars are cheaper and eventually they move production to the assembly line. Ideally the car would occupy an entire assembly line for a year. But some cars don't sell that well, so they also have to retool the line at least once to make a different car.

Retooling is expensive, so they try to avoid it by things like working twenty-four hours on another line instead. Or they build ahead so that they don't have to produce more of the current car after retooling.

Again, my point is that insurance and cars are similar. The first customer is really expensive and the second, third, etc. are almost as much. You need thousands if not millions of customers to amortize the pooled risk (insurance) or the design and setup costs (cars). Software makes this more so, as the marginal costs get much closer to zero.


"When a car company makes a new vehicle, the 10th car costs the same to manufacture as the 1000th." You're saying that there are no economies of scale? This doesn't really make sense. The cost of producing a marginal unit once an assembly or other manufacturing process is already set up is going to be less than setting up the manufacturing process in order to produce one unit.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.