Revenue is only part of the equation

In 2016, venture was all about D2C. Every mattress, suitcase, shaving razor, and gift box company was a unicorn because they had so much revenue. And the market was valuing them with SaaS multiples because they had an app.

The second part of the equation

The problem then was the same problem we’re seeing now: just because you have an app, does not mean your margins are SaaS-like (70%+). And if your margins are not SaaS-like, your multiple shouldn’t be either.

Taking a step back, company valuations are a function of two things:

• Your revenue (and specifically your revenue growth)
• Your multiple (which is generally a function of your margins)

The market is obsessed with revenue because it’s an easy number to graph. And we often ignore margin because it’s the least risky part of—and therefore the last thing to optimize when—building a software company:

1. Prove the need (most risk)
2. Capture demand (medium risk)
3. Optimize margin (least risk, for software)

It’s low risk because software sells electrons, and electrons are fundamentally cheap. However, if your app sells mattress, suitcases, shaving razors, or anything else with a real margin, your ability to optimize your margin does not benefit from the same fundamentals.

That is what happened in ~2018. Eventually, the market realized that D2C was not SaaS and the result was a lot of downrounds and M&A.

The bet on AI

To date, most AI companies have horrible—not very SaaS-like at all—margins, but because they have apps and sell electrons, the market doesn’t seem to care.

In fact, many of these companies go so far as to subsidize tokens because they believe it is an accelerator to getting usage.

Underlying this all, the massive bet we’re taking is that intelligence will commoditize and companies will be able to return to SaaS-like margins.

So instead, we’ve obsessed over these AI apps and that they’ve grown faster than we’ve ever seen. We’ve ignored the fact that they’re selling dollars for dimes because we assume that can be optimized like any other piece of software.

Saying the quiet part out loud

Recently, it’s starting to seem that AI tokens are not commodities, and actually have real margin because users expect the most cutting edge model (as opposed to some local, cost-optimzied model).

That’s making it feel a lot like 2016.

It seems like we may have created something valuable. Something worth paying a lot of money for. But just misvalued how much of that the company can capture.

If that’s right, the result of this will be one of two things:

1. We’ll re-adjust the multiple for AI companies
2. The AI companies will stop internalizing the cost of their tokens (preferring for you to bring your own key or something like that), and we’ll see the fantastic revenue curves come back to earth as we re-adjust reporting

Either way, I hope we find a way to correct gracefully, because the alternative is another 2016. Except in 2016 the margins weren’t negative.