ROI on AI

Sooner rather than later, this metric is going to weigh down on investors rushing in to cash in on the AI boom. Search went through similar panic syndromes in its earlier phase until it switched to advertising supported model and the rest is history. At the moment, it is not clear if that is an option for the various AI agents or products that are available most of which offer free versions with the premium version carrying a price tag.

Stickiness

This was the central factor in the era of internet for marketers and of course makers, because advertising revenues depended on this aspect – greater the stickiness greater the potential for advertising revenue, subject to other parameters. However, the same internet also ushered in a new generation of users who became accustomed to not paying for what they were using. When some companies began resorting to pay walls, users turned to other free resources. Yahoo was probably the only company that made money since its beginning because it embraced a transaction fee model.   

The question for AI companies then turns into one of getting users to stay with them even when users are asked to pay for the premium version. That, in turn, is a function of how addicted to the agent users become, that they are willing to pay for the premium version. And that, in turn, is a function of how users use AI – just simple stuff that can be addressed by free versions or stuff that will need the premium priced version. AI companies will have to convince users that they (users) cannot do without that specific AI product. The question is – who will be such users?

Users and users

This question was not faced by search engines because advertising helped them make boat loads of money, but AI agents/products is in a different category, because it costs way more to make them than search engines. Advertising doesn’t seem an option, so far. While using search engines was easy, that might not be the case with AI agents, if users want to get more out of it. You might have seen/read several advertisements telling people not to use AI the way they used search engines. In a fundamental sense, AI companies are also betting on users becoming more intelligent in the use of AI to extract what they want. One thing is certain: from a marketing perspective there will be a lot of learnings in understanding changes in user behaviour in relation to technology.    

In an article in The New York Times titled ‘This Is What Convinced Me OpenAI Will Run Out of Money’, Sebastian Mallaby, senior fellow at the Council on Foreign Relations, addressed the question of stickiness. According to him, “This lack of stickiness is most likely temporary, however. At some point in the not-so-distant future, a model will probably know its user so well that it will be painful to switch to a different one. It will remember every detail of conversations going back years; it will understand shopping habits, movie tastes, emotional hangups, professional aspirations. When that happens, abandoning a model might feel like a divorce — doable, but unpleasant”.

And this has to happen in sufficiently large numbers for it to become a profitable business. Until then, cash will just get spent without a quid pro quo. According to a report by The Information, OpenAI will burn $115 billion through 2029.

Sizes of investments

Meanwhile, money is being invested (or promised to be invested) in incredible amounts. As I have repeatedly said, data centres are central to everything in modern technology.  In another tech media reporting, OpenAI says it is committed to spending $1.4 trillion to build data centres roughly equating to 30 gigawatts of data centre capacity.

Moody’s Rating Agency (rival to Standard & Poor) has estimated that “at least $3 trillion will flow into data-center-related investments over the next five years”, as reported by Yahoo Finance. The report mentioned that “Six US hyperscalers – Microsoft Corp., Amazon.com Inc., Alphabet Inc., Oracle Corp., Meta Platforms Inc., and CoreWeave Inc. – are projected to invest $500 billion in data centers this year alone as capacity growth continues. While banks will maintain a “prominent role” in providing financing, Moody’s noted that other institutional investors will increasingly participate alongside banks due to the enormous capital requirements”.

What such growth in data centres mean for the use of resources is another vital question that I have addressed in another article titled ‘Sustainable AI’.

Open persuaders

I have modified ‘Hidden Persuaders’, Vance Packard’s classic study of advertising because everyone is screaming at you that you will be left behind, that you will become a relic if you didn’t wholeheartedly embrace AI, especially their version of it. The key question is this – which kind of users will help turn AI into a profitable business? Or what kind of uses or applications?  Mallaby refers to two diametrically opposite view points by two highly regarded institutions – MIT and Wharton.

As Mallaby observes, “The next big thing will be agents: The models will fill digital shopping baskets and take care of online bills. They will act for you”. Even as AI models are becoming more and more sophisticated, the question is always – will it make money for the makers of AI?

According to a MIT study last July by MIT NANDA, the answer is NO. Hardly surprising that investors got spooked. However, a Wharton study in October 2025 affirmed exactly the opposite after talking to 801 US companies.

MIT or Wharton?

Is anyone qualified to take a call? Your guess is as good as mine. No one can escape this question though. Wall Street will keep pouring in money as long as it is convinced that the story is credible and will have  a happy ending which will continue. Then it will become a story of IPO as an exit option for investors to let in new investors who believe in the story. The oscillation of values will then start.