A VC’s Guide on How to Spot an AI Bubble

A looming AI bubble may be about to burst. If you know where to look, the telltale signs of a correction are everywhere.

While it could be too soon to say if AI is the “fifth industrial revolution” or a great way to lose money, onlookers like Nnamdi Okike, 645 Venture’s founding partner, use a heuristic to distinguish froth and foolery from game-changing advancements in AI.

“One downside of the size of rounds that are happening, and how fast these rounds are occurring, is that investors are perhaps missing this idea of the business model quality,” Okike told Business Insider.

The firm backs startups from pre-seed to Series B, with notable investments including Meridian — an AI-enabled fintech platform for private equity deals — 645 led its $7 million seed round in June. Last year, it also led a $31 million Series B funding of Setpoint, a lending infrastructure company.

Among the investor class, many feel like they have seen this all before, as the eventual conclusion of the AI boom has drawn comparisons to the scammy vibe of crypto — which is once again having a moment — and even the dot-com bust that sparked the dawn of the internet age.

“It’s very hard in these early days to really understand what’s a sustainable business model that will have attractive long-term margins, that is defensible, and won’t necessarily be a race to the bottom on pricing.”

Okike walked me through his investment criteria, and whether we’re already in an AI bubble bound to pop.

This Q&A has been edited for clarity and length.

Talk about the firm’s investment ethos, notably how it is or isn’t investing in AI.

The most recent deal I was involved in was with Arbor, which is software that effectively replaces operational consultants.

It’s an example of an AI company that provides business intelligence. Historically, businesses might have had to go to a high-priced consulting firm to do certain business intelligence projects, rather than using software in real time.

We’ve shied away from extremely capital-intensive businesses; For instance, we haven’t invested in any of the broad large language model companies.

We’re looking at seed rounds where we can invest and buy a 10% stake in a business for $2 to $3 million. Even at Series A, we’re looking at rounds that might be $10 to $12 million, where we’re valuing them at anywhere from $40 million to $60 million post-money.

Related stories

We’re not investing in companies that are raising $2 billion Series A’s.

Can you elaborate on why we’re seeing such high valuations of some AI companies?

Companies are skipping multiple steps and raising rounds of a size much more commensurate with those of companies that historically had a truly established product-market fit.

When you start to lose that, you have to prove so many more things post-money, and you often are doing it without the scarcity and capital efficiency that’s needed.

I’m a big fan of the quote, ”Necessity is the mother of invention,” which forces you to make trade-offs. It forces you to say “I can only spend X on engineers,” or ‘I can only spend Y on go to market.”

If you can’t do that, it’s often evidence that one of your assumptions is wrong, your execution is bad, or the market forces you to change. That’s a good thing in the context of entrepreneurship.

When you move out of that system, it screws things up to where we are now, which is very large rounds that are much more consistent with what you’d see at super late pre-IPO, showing up at Series A or earlier.

This makes it more likely that companies raise down rounds and are wasteful in how they use their money. A hidden variable is also the quality of revenue.

Is that high-margin revenue? Is that defensible revenue? Whether it’s legal AI, financial AI, or developer tools, you’re seeing dramatically inflated rounds at every stage.

We’ve written about special-purpose investment vehicles, SPVs, and SPACs contributing to fears of an AI bubble. Why might these kinds of investments indicate that?

When you get into blank check SPACs, or SPACs that are raised without having a really clear focus on what type of company you’re going to buy, I think investors are investing somewhat blindly and trusting the owners to figure it out. Generally speaking, the last kind of SPAC bust group proved that out.

Let’s go back to why SPACs exist: there are companies designed to purchase attractive assets where there is either an inefficiency in the private markets, being able to invest in them or where the public market is a better place for those companies to exist.

Some examples of SPACs that were good investments in the last wave were DraftKings and Hims. In the case of DraftKings, the SPAC structure enabled multiple businesses to merge, which was a good thing because they were complementary.

Related stories

They go wrong when there’s a disconnect between how the private and public markets may view the business.

Because the public markets are quite frothy, my guess is SPAC investors might be thinking, ‘Ok, there’s a lot of demand for AI companies, so we’ll enable public market investors to get access to theoretically attractive AI assets, and those will trade at a premium.’

When we see AI companies trading at 50 or 100 times revenues, or raising rounds at that, what’s the end game? If you’re eventually valued in the public market at five to 10 times revenue, you have to grow dramatically to overcome that valuation reduction.

Besides the means of investment in AI, what other indicators are there that we’re headed into an AI bubble?

There are a few general signals of bubbles, and they’ve been relatively prevalent across the different hype cycles in tech. 1) a rapid increase in valuations, 2) the increasing frequency of new funding rounds, 3) investors investing in the greater fool theory, or investing based on expectations of what other investors are going to do, versus fundamentals, and then 4) increases in prices based upon fundamental metrics, so things like price to revenue, price to earnings, etc;.

OpenAI’s valuation went from $300 billion in the last round to $500 billion in their tender offer. It’s quite rare for a private company to be valued at even $100 million. Now you have companies valued at much more than that and increasing their valuations within periods of months.

Since the biggest companies are valued at a much larger percentage of the public markets, they can influence the performance of those markets over the long term. If you look at the Magnificent Seven, they’re not valued at extremely high multiples of earnings, but you do see their capital expenditures going up dramatically.

It’s estimated that $560 billion has been invested in capex this year, versus the roughly $35 billion of revenues generated by that spend. OpenAI is about $10 billion of this, which is them basically buying Microsoft Azure and services. So you discount that a little bit because Microsoft is selling that at cost. There’s a massive disconnect between capex spending and the revenue generated by that spending.

Given the big concentration of the public market in these companies, changes to their spending may cause a chain reaction.

You also see certain public companies trading at very high multiples of revenue. One is Palantir. It’s trading at over 100 times its revenues, which is quite extreme and rare. Palantir’s a great company, but when you start to look at those multiples, you say, ‘Hmm, that’s reminiscent of previous bubble periods.’

If your conclusion about an AI bubble is right, what might we start to see, and what could be the fallout for Big Tech?

Often, in market cycles, there’s a correction, and the corrections usually lead to companies tightening their purse strings, focusing on profitability and high-return investments.

The irony is that after the pandemic-driven 2021 tech run-up, there was a correction. There was this focus on profitability and reduction in spending in both private and public markets, and this idea that these unicorns were having to raise down rounds, get profitable, and focus on efficiency, but then you had AI show up, and much of that was thrown out the window.

You didn’t really have a full correction; some of which is still shaking itself out. I suspect something similar may occur after this boom in AI. I guess the question is, how significant or permanent will that be? And will companies at some point be forced to really tighten the purse strings?

This would mean much less investment in capex for the Magnificent Seven, likely smaller venture rounds, and companies would have to be a lot more efficient in how they’re training their models.

What’s interesting is that you had this datapoint from Deepseek. The market reacted very dramatically to this one-off company that was able to train its models for a fraction of what the US LLMs were training their models.

The market seems to have forgotten about it, and companies just kept investing at the rates they were. It’s a strange outlier event.