It's common to observe that the number of publicly-traded companies in the US, and elsewhere in the world, is in secular decline. But that long-term decline trend has a few epicycles that throw things off—the actual trend, looking at the cohort of companies-that-would-typically-go-public-circa-the-1980s, is actually worse than popularly understood, because there have been two bursts of public offerings specifically fueled by investors' willingness to back pre-profitable companies (in the 90s) and pre-revenue companies, some of which were barely an idea and an R&D line item (in the SPAC boom).

It's long been the case that companies consume more capital than they produce early on. If that weren't true, you would expect the financial system in general to be a lot smaller, since everyone would fund growth from operations. That's a world where fewer people spend all day, or all their life, looking at spreadsheets. But it's also a world where economic activity is more weighted towards mature companies, instead of emerging startups.

It's a good thing that public market investors got accustomed to backing companies before they could turn a profit. But it's worth looking at why: if you examine the IPOs of the earliest generation of software companies, you'll find something unfamiliar: a growth company that waits until it's profitable to go public! Consider the first Adobe annual report, or Microsoft's prospectus. Where's the red ink!?

One possibility is that companies and founders were a lot better back then, and that investors were less forgiving and simply insisted on a positive P&L. But that's probably not it. Instead, the nature of the business changed—in particular, it changed in a way that made it more strategically optimal for companies to lose money for a while in order to maximize profits early.

Early PC software companies made their money one product at a time, generally tied to new releases and hardware upgrade cycles. More PCs in use meant more computers that needed to run MS-DOS, Quattro, Lotus 1-2-3, etc. But these products were generally one-off sales. That made early software economics look more like the movie business: any given title's profits would be unpredictable, but the average result needed to be positive because there wasn't much left after that. (The operating system business was different, with per-computer licensing fees that meant a much smaller amount of revenue per user, multiplied by a larger number of users. But in this case, there was still a connection to the number of units shipped, and it was harder to push through a price hike without having a new OS as an excuse to do so.)

Hardcore Software covers the period when this started to change: when Microsoft realized that the company with the dominant operating system and developer tools would be able to make and ship the best apps. And that this changed the economics of software products. Instead of looking at each product on its own, you'd want to measure the complementarity, which could easily be cross-platform: not only did Windows make it easier to sell office software, but if the standard office software worked best on Windows, then Windows was an even better default operating system.

That doesn't just make software a better business, at least for the winners. (It made it a much worse business for runners-up like Borland.) But it also made it a very different business: expecting every product to pay its own way leaves money on the table when there are network effects across different products.

In a way, the transition to SaaS was just part of a natural continuum: people were obviously going to buy a lot of software, and they'd sometimes pay for upgrades, but that was a problem: if different people run different versions of the same software, incompatibilities will creep in, and building not just the next version but backwards compatibility with all previous versions means that the cost of each new product scales partly as a function of the cumulative install base of the previous versions. A better option, especially once most homes and offices had reasonably fast connections, was to switch to a recurring model, and continuously ship updates, either by letting people download product updates or, for browser-based products, just telling them to hit refresh.

Once you've turned a product like Office from a $499 one-time purchase to a $100/year subscription, your economics have changed in a big way: now, you care about spending money not to create current-period revenue, but to create the most valuable possible stream of future revenue. And if you're acquiring customers for, say, $200, that $100/year annuity is a pretty good deal unless churn is awful.

But when that happens, the growth phase is also a phase where companies deliberately try to lose money. If you can turn $200 into maybe $600 in present value, you want to run that operation as many times as you can, even though every incremental sale produces an additional $100 loss in the first year. Companies devote a lot of thought and cleverness to scaling their money-losing as effectively as possible, recruiting armies of salespeople in order to do as many negative-profit transactions as possible.

Because, of course, if you matched the economics, amortizing the cost of developing and marketing the software over the life of each contract, you'd see a profitable business instead. Capitalizing marketing is tricky stuff; AOL got in trouble for this in the 90s, and got themselves in trouble by overspending on growth even as their economics deteriorated. But it's true of any company that realizes value over time.

This usually goes without saying, but if you don't have it in the back of your mind when looking at early companies, it's easy to make mistakes. This piece is an incredibly breathless take on how OpenAI is careening towards disaster, but strip away the emotional language and a summary is: "OpenAI is an unusually large startup, with a typical 12-24 months of runway. Their backers believe that their future prospects are bright, but there is of course room for disagreement there—after all, if everybody felt that way, nobody would have had to start OpenAI!" This broad description probably applies to about 95% of venture-funded tech, by market value. Almost every growth company is, by design, going to run out of money in a year or two unless it raises more, in the same way that most flights are on track to run out of fuel and crash unless the pilot lands them in the meantime. It's entirely possible that their unit economics won't work out eventually, and that they really will run out of cash (and get acquired, not go bankrupt). But that question hinges more on some admittedly speculative questions about the future course of AI and the ability of individual labs to develop durable competitive advantages. Those are interesting questions, and worth discussing. And, in fact, the phenomenon that the most successful companies in tech lose money for longer than they used to is also worth talking about. But the goal of understanding that is to be in a position to treat any one company's lack of profits, and short runway, as a commonplace observation.

And you can flip it around. What would it mean if OpenAI were to turn a profit? Going by their current model, it would mean that they'd cut some costs a fair amount, though you don't need as many seven-figure researchers if there aren't new models being researched. But it would also mean that AI capabilities leveled out at roughly their current level—about as fairly smart people on some tasks, like drafting letters and writing bits of code; not great at other tasks, like knowing when to switch from reasoning in text to reasoning in code (though that, too, makes them endearingly humanoid—a lot of figures get thrown around by people who clearly aren't in the habit of running some numbers and asking what else has to be true for what they're saying to be true); it would replace some customer service work, speed up some legal and programming work, replace some interns and outsourced call center workers, and otherwise be kind of a bust. To an AI optimist, the worst thing OpenAI can ever tell them is that it's tried to turn a profit. (To an OpenAI investor, that's second-worst: the worst is to try to do that and then fail.)

For a growth company, profitability and free cash flow are bittersweet. They mean that the company's reaping financial rewards, that it doesn't have to listen to investors as much, and that it's all grown up—but that also means that it simply can't find enough good places to put the next dollar it earns, and has no choice but to return them. The longer companies spend in this state, the harder it will be to make the switch, but in the end, the reason investors keep providing that next 12 months of runway, over and over again and at escalating valuations, is because they expect the growth trend to end some day and for the company to have so few good ideas that the best one is buying back stock.

Disclosure: Long MSFT.

Read More in The Diff

We've covered the optimal cash burn rate question in a few places in The Diff. For example:

• We've looked at startup swag, and why money-losing companies spend some of their limited runway slapping their logo on t-shirts and mugs.

• We've examined how equity's value partly comes from being an option on future reinvestment.

• In "The Other Moviepass Economy," we looked at the transition to subscription-plus-transaction models, and in Costcoing consumables ($) we take a closer look at where this model works extraordinarily well (who knew that food delivery and 3D printers have such deep economic similarities?).

• And we've considered how abundant later-stage funding changes the math for early-stage valuations ($).

Join the discussion!