Is the AI bubble about to burst or just getting started? The smartest people in the room can’t agree, and the reason why tells you more than any prediction.

In the same week this June, two things happened.

Ray Dalio, the billionaire who built the world’s largest hedge fund, went on television and said the AI market is showing the signs of a bubble that will eventually burst. A few days later, the CEO of Hewlett Packard Enterprise looked at his order book, said “there is real demand,” and did something almost nobody does in June: he issued financial guidance for 2027. Not because he was nervous. Because he was that confident.

One of them sees a cliff. The other sees a runway. They’re looking at the exact same industry.

That’s the strange thing about the AI bubble in 2026. This isn’t a fight between people who understand the technology and people who don’t. It’s a fight between people who understand it equally well and still land on opposite conclusions. So before you pick a side, it’s worth understanding why both sides are, in their own way, completely right.

First, what does “AI bubble” even mean?

A bubble isn’t “a thing that is expensive.” A bubble is when the price of something detaches from the value it actually produces when people are buying not because of what it earns today, but because of what they’re convinced it will earn tomorrow.

By that definition, the question for AI is brutally simple: Is the world spending money in line with what AI is actually delivering?

Hold that question in your head, because the numbers are about to get uncomfortable.

The case that the AI bubble is real

Start with the gap. Gartner projects worldwide AI spending will hit roughly $2.5 trillion in 2026. To put that in perspective, that’s larger than the entire annual economic output of most countries on earth.

Now look at what’s coming back the other way. OpenAI, the company that is the AI story for most people, is running at somewhere around $25 billion in annualised revenue and is reportedly on track to lose about $14 billion this year, nearly triple last year’s losses. The company has signed infrastructure commitments measured in the trillions, against revenue measured in the tens of billions. The math only closes if the future arrives exactly on schedule.

Then there’s the most quietly devastating statistic of all. A widely-cited MIT study found that 95% of enterprises report zero measurable return on their generative-AI investments. Not “small returns.” Zero, measurable. Most of the companies that bought the dream haven’t been able to prove it paid off.

This is the heart of the sceptics’ argument. As Goldman Sachs’ head of equity research, James Covello put it on the bank’s own podcast, it all comes down to one question: do the companies using AI actually make or save money? If they do, the technology fulfils its promise. If they don’t, the spending was a story we told ourselves.

And then there’s the part that makes seasoned investors genuinely nervous: circular financing.

Here’s the loop, stripped down. A chipmaker invests billions in an AI lab. The AI lab uses that money to buy chips and cloud capacity from the same chipmaker and its partners. The cloud provider then borrows more money to buy more chips. Money goes out one door and comes back in another, and at every step, somebody books it as “revenue” or “demand.” By 2026 estimates, more than $800 billion in deals are tangled up in arrangements like this, Nvidia, OpenAI, Oracle, AMD, Microsoft, all appearing on multiple sides of the same transactions.

The danger isn’t that it’s illegal. It isn’t, and it isn’t even hidden; anyone can trace the deals. The danger is that it can manufacture the illusion of demand. When the companies selling the shovels are also funding the people digging, it gets very hard to tell how much of the gold rush is real.

The economist Ruchir Sharma summed up the bear case with what he calls the “four O’s.” He argues AI is now flashing red on all four: overinvestment, overvaluation, over-ownership, and over-leverage. Wealth is concentrated in a tiny handful of stocks. The market’s price-to-earnings measures have pushed into territory last seen right before the dot-com crash. And after years of sitting on cash, Big Tech has started borrowing heavily to keep the spending going.

No wonder Deutsche Bank described 2026 as the year “1999 meets 1990.”

The case against the AI bubble

And yet.

The single most important word in this whole debate might be the one the HPE chief executive used: demand. Because while the headlines scream “bubble,” the people actually selling AI infrastructure are reporting the best quarters of their careers. Dell’s stock leapt more than 30% on a single earnings report. Cisco, HPE, Nvidia, and company after company are posting numbers that don’t look like a market running on fumes. They look like a market that genuinely can’t build fast enough.

That’s the Bulls’ first point: this isn’t 1999, because in 1999, the revenue wasn’t there. Pets.com had a sock puppet and no profits. Today’s AI giants are, for the most part, real businesses with real customers writing real cheques.

Their second point is about who’s paying. The dot-com bubble was inflated by speculative startups burning through borrowed money. A large share of today’s AI build-out is being funded by some of the most profitable companies in human history, out of their own pockets. A bubble funded by Microsoft’s cash flow is a very different animal from a bubble funded by a stranger’s credit card.

And here’s the kicker that should make even the sceptics pause: the betting markets agree with the bulls. On Polymarket, where people put actual money behind their predictions, traders are pricing only around a 16% chance of the AI bubble bursting by the end of 2026. The crowd that’s literally wagering on a crash mostly doesn’t expect one.

The dot-com lesson the AI bubble debate forgets

Here’s where the smartest take lives, and it’s not in either camp.

Everyone reaches for the dot-com comparison, but they almost always use it wrong. They use it to mean “it was a bubble, it popped, the end.”

That’s not what happened.

The dot-com era was absolutely a bubble. Trillions in value evaporated. Companies died by the hundred. People lost their savings. And it also built the internet we live on today. All that “wasteful” overinvestment laid thousands of miles of fibre-optic cable across the world, cable that sat dark and unused when the bubble burst, and that a few years later carried YouTube, Netflix, Spotify, and every app on your phone. The crash was real. The transformation was also real. Both things were true at once.

That’s the frame for AI in 2026. The bubble talk and the revolution talk aren’t actually contradictory. It is entirely possible, maybe even likely, that:

• A painful correction is coming, where overhyped valuations snap back, and a lot of “AI-powered” companies with no real product quietly disappear, and
• The infrastructure being built right now, the data centres and the models, becomes the foundation for the most important technology platform of our generation.

The dot-com bust didn’t kill the internet. It killed the froth around the internet and left the substance standing. Amazon fell more than 90% and then became Amazon.

So “is AI a bubble?” might be the wrong question. The better question is: when the froth clears, what’s left standing, and is it the thing you bet on?

What this actually means for you

If you’re an investor, a founder, or a business leader watching all this, the temptation is to do something dramatic, pile in before you miss out, or pull out before it all collapses. Both are reactions to the noise.

The signal is quieter, and it sounds a lot like common sense:

Separate the technology from the trade. AI being genuinely transformative and AI stocks being overpriced can both be true. Don’t let your view of one decide your view of the other.

Chase ROI, not vibes. Remember that 95% statistic. The companies that win the next five years won’t be the ones that spent the most on AI; they’ll be the ones who found the specific, boring, measurable place where it actually saves time or makes money. Pick the use case you can prove. Ignore the use case, you can only post about.

Build on fundamentals, not financing loops. When the correction comes, whenever it comes, the things that survive will be the ones solving a real problem for real customers. That was true in 2001. It will be true here.

The AI bubble debate isn’t really a question of whether the technology matters. It clearly does. It’s a question of how much we’ve priced in before the proof arrives and whether you’re standing on the part that lasts, or the part that pops.

Dalio and the HPE chief executive will both, eventually, be proven right. There will be a reckoning, and there will be a revolution. The only thing that’s genuinely up to you is which side of that you choose to build on.

What’s your read bubble, boom, or both? We’d love to hear how you’re thinking about AI in your own work.

Read More Here