The expectation that modern AI tech will find a home in every part of our lives is pandemic. Fittingly, startups and investors are working overtime to build and fund new technology companies to either create or implement new AI tech. Major rounds are often in the headlines, and startups are building at breakneck speeds to stay ahead of both the technology curve and the largest tech companies that have their own AI strategies.
But despite all the enthusiasm, there’s a niggling detail that deserves our attention: AI startups often have worse economics than most software startups.
The fact that Anthropic, a leading AI startup that has raised billions of dollars, reportedly had gross margins of 50% to 55% last December underscores the costs of building and running modern AI models, and hints that AI-focused startups have a different valuation profile due to the sheer expense of all that computing power.
The Exchange explores startups, markets and money.
Read it every morning on TechCrunch+ or get The Exchange newsletter every Saturday.
Revenue quality is partially predicated on gross margins — revenue less costs of goods sold — and the better those margins, the better the revenue, all else held equal. Startups have long depended on revenue quality as an explanation for their impressive losses during their scaling years — yes, startups consume lots of cash, but the revenue they generate is pristine in terms of quality, and thus worth quite a lot.
This is, among other reasons, why software companies are frequently valued on a multiple of their revenue instead of their profits. When gross margins are high, strong revenue yields oodles of gross profit. Investors like that. But that’s not a valuation model that you can apply to a company that’s, say, selling groceries.
The conversation around AI gross margins is not new. Back in 2020, venture firm a16z argued that AI startups would have lower gross margins due to “heavy cloud infrastructure usage and ongoing human support.”
According to The Information’s new report on Anthropic, those are the exact factors affecting the company’s gross margins, so we can infer that the overall economic profile of AI startups has not changed much since 2020, despite advances in AI technology. Sure, generative transformers were transformative, but AI still requires lots of compute cycles, and models are growing much faster than computers can keep up, so it’s still an expensive game.
Back in 2020, I wrote the following based on a16z’s argument:
If a16z is correct about AI startups having slimmer gross margins than SaaS companies, they should — all other things held equal — be worth less per dollar of revenue generated; or in simpler terms, they should trade at a revenue multiple discount to SaaS companies, leaving the latter category of technology company still atop the valuation hierarchy.
Since then, we’ve seen a venture bubble form and pop, and the value of SaaS companies also bubbled and popped similarly. A company that was once worth 20x revenue in 2021 — ARR, if you will — might now be worth 7x or less.
AI, however, remains as scintillating a category as ever. As a result, many startups in that field are seeing their revenue multiples stretch to Icarian heights. Indeed, as I wrote in our TC+ Today newsletter the other day:
Perplexity.ai recently reached annual recurring revenue of $6 million, which is double the revenue it recorded in October 2023. That’s the sort of revenue increase venture investors covet. It’s no surprise, then, that Perplexity just raised $73.6 million at a valuation of $520 million.
Now, that valuation yields a really high revenue multiple (87x ARR), and is reminiscent of the valuations we saw in 2021.
See the problem? If SaaS valuations have retreated sharply, and those companies have much higher gross margins than AI-startups can hope to generate, it’s odd to see investors’ wagers flying into the pot at such high prices.
Of course, there’s some real logic here besides the usual FOMO: AI startups are growing very quickly, and startups that grow faster than their peers earn valuation premiums. And if this fancy tech really is going to shake up the whole world, well, who cares what price you have to buy in at? You can still make 100x your money! Right?
Maybe. The AI vs. SaaS conversation is muddled by the simple fact that lots of AI companies are SaaS companies. Which is why, of course, when we digest SaaS metrics, we tend to bucket them into subgroups so that we can do more effective analysis.
To honor that work, the bottom quartile of companies in the Bessemer Cloud Index are posting gross margins of about 69% today. Put another way, every AI startup that is stuck reporting gross margins in the 50s and low 60s will find itself in the basement of SaaS company lists. That means AI startups that do scale will likely generate less cash flow than a pure SaaS startup of similar scale, so should be valued more conservatively.
This is all very interesting and worth keeping an eye on. I presume we’ll return to this conversation every few years. See you in 2027!