Devina Mehra: Whether there’s an AI bubble or not is a big question—here’s the lowdown on what’s brewing

Yes, I do have concerns. Especially if one looks at the US market. Its move has been very narrow. It has basically been a tech move, and even within that, just a handful of companies have been powering the whole rally.

Even globally, the whole equity market move has had a contribution of 70% from technology stocks. Among Nasdaq shares, the biggest returns have come from companies with either no revenues or no profits. And there is this mindset around new technologies, an almost irresistible urge to ‘go chase the next new thing.’ But it’s often just a mirage. Because the real questions are:

How long will this new technology take to succeed? Who will actually succeed? Or, to be specific, which particular technology and which company will succeed? And, most importantly, will even that company make enough of a return on capital employed?

Let us go back to the dotcom boom. Back then, there was this thinking that if the internet will be used by everybody in the world—which turned out to be true—then you cannot go wrong investing in companies that are building the infrastructure for it.

The internet is carried across the world on undersea cables. There was a company called Global Crossing that laid many of these cables. These are still in use, but the company went bankrupt more than 20 years ago. So it is not only about whether a particular technology succeeds. Will it earn a return on the massive amount of money being invested? That’s the question.

Coming to AI, large companies in the AI race right now, such as Meta, Google and others, are hiring individuals and buying very small companies with just one or two people at eye-popping figures like $100 million and $500 million.

They are also setting up huge data centres. Now what is a data centre? It is real estate. It is a very substantial investment in buildings and equipment, which depreciate very quickly. As for their running costs, they require a lot of energy.

The cost of being in the AI game is rising. And the dilemma is that if you run a technology company, you cannot afford not to be in the game. But, again, the real question for any investment is: Are you really going to make good returns on it?

Here is one damning fact: when schools and colleges shut down for vacations in the US recently, the usage of many AI models dropped by 50–70%. I am sure if we map it onto school breaks in other regions as well, this percentage will rise even further. The point? These models are mainly used by students for their assignments. That’s not exactly a high-paying user base. Moreover, most AI companies are losing money, and not just on an aggregate basis. With every incremental subscriber, their losses go up, not down.

An additional danger is that this over-investment cycle could spill over to the world of finance. This is visible in two things.

One is the debt being raised, not just by existing mega players, but a number of much smaller players. A New York Times article on 10 November titled ‘Debt Funding Racheting Up Risks of AI Boom’ details some of this debt-raising. For example, Meta has committed to buy $14 billion worth of computing power from a company called Core Weave, which in turn is financing about 55% of its investments via debt. There is a similar arrangement between Microsoft and Nebius, a startup.

By some estimates, the total debt taken on for data centres will cross $1 trillion by 2028. Loads of debt has already been taken against buildings and equipment as security, most of which can be expected to depreciate fast, leaving that much less value to be reclaimed by lenders if it comes to that eventuality.

Then there are also a bunch of financial tangles that you may have seen in the form of infographics. For example, Nvidia and Microsoft funding OpenAI, which gives them business in turn, either directly or through Oracle.

This crisscross of lines between companies looks like a spider’s web. But net of all the details, it is a case of capital funding by bigger companies like Nvidia, Meta or Microsoft coming back as revenue for them. So negative fund flows show up only on balance sheets, while boosting revenues and profits in income statements.

In simple terms, Company A invests in the debt or equity of its customer B. This capital infusion is used to buy goods from A, which boosts its revenues and profit. This type of circular financing is a red flag in general.

But for now, the AI music is still playing and, as Citibank’s then head Chuck Prince said before the 2008 blow-up in America’s mortgage market, “As long as the music is playing, you’ve got to get up and dance.”

That, in some sense, is the dilemma of all tech companies in the world. While the AI race is on, you cannot opt out of it without protests from your stakeholders, even if the race veers into risky territory.

Let me end with one of my favourite examples of investing in successful technologies. Which were the two technologies of the 20th century that fundamentally transformed how human beings live? One was the automobile and the other aviation. Both sectors have been graveyards for companies. The two have largely been terrible industries to invest in, even though these technologies succeeded so dramatically.

The author is founder of First Global and author of ‘Money, Myths and Mantras: The Ultimate Investment Guide’. Her X handle is @devinamehra.