Will AI Follow Enterprise SaaS or Clean Tech?
The 10,000th post on whether AI is in a bubble (but with cool data and charts!)
This quick post was inspired by my Twitter/X exchange with Dan Gray and Truebridge Capital’s Mel Williams’ two-minute spiel on a recent This Week in Startups episode, where he discussed AI’s similarities to the green tech bubble in the mid-late 2000s.
In the podcast, Mel makes two cases for why many AI companies may fail, as it relates to clean tech, and why many AI companies may succeed, as it relates to enterprise SaaS.
Fail (Clean Tech):
“High capital requirements, long time required to build a scale genuine scientific and technical risks that weren't well understood by the venture industry. Low margin businesses, regulatory risk, competition from substitute products. A lot of that sounds familiar when you think about AI today.”
Succeed (Enterprise SaaS):
“And then you look at why enterprise SaaS succeeded over a very long period of time. It's because you have inherently scalable companies. You have recurrent revenue models.
You have extremely high gross margins. And a lot of that sounds familiar when you think about AI. So, we do think that AI has attributes of both”
He makes accurate statements for both claims, but if I were an investor, I’d be more uncomfortable with AI’s similarities to green tech than bullish on its similarities to enterprise SaaS.
Many AI companies have massive capital requirements, as proven by these gargantuan “seed” rounds, just to purchase computing power. Look no further than Inflection.ai. All of that capital is spent on creating low-margin, non-defensible businesses. Just like green tech spurred thousands of solar panel companies, we’re seeing thousands of AI co-pilot companies, with both types of businesses being very non-differentiated for the most part.
Sure, you can sell AI software as an enterprise SaaS product and generate meaningful revenue, but what’s to say OpenAI doesn’t just copy that with its ridiculously large budget and intelligence moat?
What’s also different about this innovation cycle is that incumbents like Microsoft, Meta, and Google directly compete with startups. Even Google has begun disrupting its own ten-blue-link business model to compete in the AI race.
People are already saying Perplexity, which just raised at a $1 billion valuation last month, may fail because Google is beginning to embrace being a true answer machine to searches, as Perplexity currently is. Perplexity was the Google disrupter, but if Google disrupts itself, where does that leave Perplexity?
Probably with no customers and a lot of disappointed investors.
How to Invest in a Bubble:
Even in the dot-com bubble, there were Amazon and Google, and in the green tech bubble, there was Tesla. Even if AI is in a bubble, it’s too remarkable an innovation not to have at least one massive winner. The problem for investors is that the winner very well may have been crowned in OpenAI, as supported by its near $90 billion valuation and deep partnership with Microsoft.
A few months ago, Founders Fund’s Brian Singerman mentioned how one of their investment theses is to avoid the hype cycles, invest in the clear winner in that cycle, and call it a day. He mentioned how Founders Fund invested in OpenAI, what he considered the clear winner, and then said they’d likely not make any more AI investments. Not too long after, Founders Fund led a round in Cognition Labs, the company behind the software engineer AI tool Devin, but those are the only two investments in “AI” they’ve made.
All new markets/bubbles follow the same trends: There will likely be one outsized winner, and the bigger the bubble, the bigger the absolute winner. Why? Capital flows to so many companies that once all of those companies fail, capital will get aggregated to the clear winners as the only place to make that bet. Amazon, Tesla, Coinbase, OpenAI—it works itself out eventually.
I’m still amazed by how wide a margin these market leaders win. Coinbase is more than twice as valuable as the next ten crypto companies combined, and Uber is three times as valuable as the next ten ride-sharing companies combined. I’ll reference this chart by Trae Stephens again to showcase just how insane this margin is.
So, if this AI wave is a bubble that gravitates towards something similar to the green tech bubble, will OpenAI be this cycle’s Tesla? It looks like it. Where I think it will be different from the green tech bubble is that Tesla was really the only big winner. As you can see, that market was a disaster.
I mean, even ZIRP couldn’t bring green tech back to the level it was in 2008. That’s really saying something. I expect this AI bubble to play out more like the dot-com bubble, where there will be a lot of failures, with some successful companies that persevere and a few massive winners.
Why It’s Too Late to Invest in AI:
While I believe OpenAI is likely the winner and investing in more AI foundation model companies is pointless, one could make the classic argument that Google was the 17th search engine and Facebook was the 100th social network company or whatever, but as I said, this market is way more reliant on capital than search and social networks. Facebook and Google built juggernauts with less than $25 million, whereas OpenAI built a juggernaut with $14 billion. You just can’t compete with those resources. This market is so reliant on capital, and a startup will have to make a god-level argument as to why they can compete with a $90 billion company that has raised $14 billion.
Thus, it’s too late.
As this chart shows, the dot-com bubble peaked in the worst year of returns for VC funds, 1999. On the contrary, the best fund vintages, 95-97, invested in the companies that created the bubble, not the ones that built during it.
I think this chart is a fantastic visualization of how venture investors should think about investing during cycles. I suspect investors who followed the Founders Fund mantra that very few winners emerge from a bubble will generate healthy returns, whereas those who invested in every AI company they could once everyone started talking about it as “the new internet” after ChatGPT’s launch will probably lose money.
I can’t wait to see how this cycle plays out. It will be a remarkable case study, once again, regarding how to invest in cycles as a venture capitalist.
I think a simple rule is to invest in the things that sound crazy to most investors but that the smartest founders are working on. Elon Musk and Sam Altman, two bonafide geniuses, founded OpenAI in 2015, not 2022 when ChatGPT came out (obviously, but you get my point). That’s the time to invest in the cycle: At the outset, when the smartest founders sound the craziest to the most people.
How do LPs manage Cycles:
I noted a few weeks ago that specialist investors outperform generalists, but I suspect that’s because the specialists hop on and ride the wave before everyone else. The specialists were AI investors in 2015-2017. Any fund starting today “to invest in AI” or “adding AI” to their thesis is almost certainly too late, so why do I hear so many LPs asking, “What’s your AI play?” to so many emerging funds? Isn’t it too late? Shouldn't they be looking for the next hype cycle? Sure, that cycle will almost certainly be supported by AI, but it won’t BE AI.
I suspect that’s the unfortunate tradeoff of raising a specialized fund that substantially outperforms: Most LPs won’t get it when the technology is so new.
In 2015, they couldn’t see what AI could be, so I assume most LPs wouldn’t give you money to invest in that category. However, that 2015 AI fund, if it got into OpenAI in 2018, for example, would likely be the highest-performing fund started in 2015.
So it’s an interesting see-saw effect. When everyone is doing it, an LP shouldn’t back that fund manager, but when it’s something specialized that sounds bizarre and hasn’t shown signs of accruing value to date, those might actually be the best funds to invest in.
It’s hard to say what that next great specialization will be. Maybe two years ago, you could say robotics, but that seems to already be too mainstream today, like space. Maybe something like agriculture tech could finally have its moment, with Ohalo leading the way for more pioneers while services like AlphaFold 3 make testing easier. Though, if I’m predicting it as a generalist myself, then that’s probably too late as well.
Best to look to the freaks for what’s next, as always.