AI could become essential to everyday life and still disappoint its investors. I think that possibility gets too little attention in arguments about whether the AI bubble is popping.

I don't expect the same outcome for every company. Some will build lasting businesses. Others may grow fast and still struggle to justify the capital they used.

The biggest risk is in the expectations: how soon AI must turn a profit, how much infrastructure it needs, and how much of its value reaches the companies paying for it. The Motley Fool's video asks the right question, but calling the whole industry a bubble groups together very different businesses. A profitable software company adding AI to a product it already sells faces one problem. A company borrowing heavily to build computing capacity for a handful of customers faces another.

The demand is real

Microsoft reported that Azure and other cloud services revenue grew 43% in the quarter ended June 30, 2026. (Microsoft's quarterly results)

That figure covers more than AI, so it can't prove AI spending pays off on its own. It does show real commercial demand in one of the businesses supporting the buildout.

So is the spending

Meta reported $32.79 billion in operating cash flow for the third quarter of 2026, against $34.91 billion in capital expenditures, including finance lease principal payments. Its reported free cash flow was negative $2.12 billion. (Meta's quarterly results)

These are companywide numbers, and one quarter can't tell us whether a long-term bet will work. They do show how infrastructure commitments can swallow the cash a successful business produces.

My concern is the gap between spending money and proving it was well spent.

When every decision makes sense and the total doesn't

A company can have good reasons to invest aggressively. Waiting could mean losing customers, talent, or access to scarce infrastructure. But when every competitor follows that logic at once, combined spending can exceed what the market ends up rewarding. Every boardroom can make a defensible call while the industry as a whole builds too much.

Physical limits make this harder. The International Energy Agency expects the largest data center operators to spend more than $700 billion in capital expenditures in 2026, mostly on new data centers. (IEA analysis) It also points to grid constraints, long connection queues, and duplicate or speculative projects in those queues.

I read announced demand with that in mind. A proposed facility is still several steps away from a working business with paying customers.

Useful isn't the same as profitable

I see another risk for investors: AI could deliver a lot of value to users while competition limits what providers can charge.

Stanford's 2026 AI Index estimates that generative AI tools gave U.S. consumers $172 billion in annual value by early 2026, often through services they used for free. (Stanford AI Index) That measures benefit to consumers, not revenue the AI companies can split. Usefulness alone doesn't tell me whether providers can collect enough to justify what they spent.

If models keep getting cheaper to run and easier to swap, customers can get more done for less money, and adoption can accelerate. Providers then need more volume, lower costs, or stronger customer loyalty to maintain returns. Technological progress can make life better for users and harder for investors.

The case for spending anyway

The strongest case for today's spending is that capacity has to exist before many new applications can. Some of that capacity will look excessive until customers find out what to do with it. Demanding an immediate profit from every project can lead to its own costly mistake: building too little, too late.

That argument still needs milestones. I'd get more confident if companies show recurring customer spending, margins that improve after computing costs, and cash generation that keeps up with expansion. I'd grow more skeptical if growth relied increasingly on fresh financing while the cost of serving customers stayed unclear.

The three tests

I keep coming back to ordinary business questions:

  • Do customers renew?
  • Do prices cover costs?
  • Is there cash left after the next round of investment?

The technology can keep improving while those answers push investors to rethink what they're willing to pay. The companies paying for AI still have to earn their money back.

This is my opinion, not investment advice.