The AI bubble is real. The technology isn't a fad.

Every week someone asks me: “Is AI a bubble?”. My answer is that the question hides two different questions, and that’s where the confusion starts. Is there a financial bubble? Probably yes. Is the technology a passing fad? Absolutely not. Both can be true at once, and it has happened before.

The signs of a bubble

Anyone who studied finance recognises certain patterns, and in 2026 we see almost all of them.

Valuations running ahead of revenue. The companies building models are worth hundreds of billions (Bloomberg titled a recent documentary on Anthropic “the $965 billion juggernaut”), with revenue growing very fast but still a fraction of those figures. The market is paying today for profits that, if they come, will come many years from now.

Huge infrastructure spending. Data centres, chips, energy: hundreds of billions a year. Ray Dalio, hardly a professional doom-monger, has said plainly he sees the hallmarks of a bubble: capital spending that current profits don’t justify.

Money going round in circles. The chipmaker invests in the AI lab, which uses the money to buy chips from the chipmaker, which books the revenue. The cloud provider funds the customer that buys its cloud. Bloomberg devoted an episode to this “money loop” with a blunt title: it feels like 2008.

Debt. As long as the giants paid for investment out of their profits, the risk stayed inside their balance sheets. Now a growing share of data centres is debt-financed, often off-balance-sheet. If expected revenue is late, the problem shifts to the lenders.

Why the technology will stay

That said, concluding “so AI is a fad” is the opposite mistake. Some facts a fad doesn’t have:

• Usage is real and massive. ChatGPT has passed 1.2 billion weekly users. Not curious downloads: people coming back every week.
• Costs are collapsing. According to Artificial Analysis, GPT-6.1 Sol solves one of their benchmark tasks for 72 cents, a few points behind the most capable model, which spends almost $6 on the same task. When a technology becomes more capable and cheaper at once, it doesn’t vanish: it spreads.
• Value is measurable in concrete tasks. Reading 1,000 invoices a month with Claude Haiku 5.5 costs under a dollar of model time. Drafting a first version of an opinion takes minutes instead of hours. You don’t need promises about general intelligence to justify these uses.

The dot-com lesson

In 2000 the internet bubble burst spectacularly. Many companies went bust, the Nasdaq lost almost 80%. Yet the internet didn’t disappear: it became the infrastructure of everything. The companies that survived, and those born afterwards on the ruins at lower prices, built the economy of the last twenty years.

I think something similar will happen with AI. Some labs won’t make it; some data centres will sit half empty; some investors will lose a lot. But the ability to read, write, analyse and code at near-zero cost won’t go back.

What it means for people who use AI

For a professional or small business the “bubble or not” question matters less than it seems. Practical consequences:

1. Don’t tie yourself to one provider. If a company gets into trouble, it changes prices or terms. Keep your processes and prompts portable, your data exportable, and know at least one alternative for every critical tool.
2. Expect unstable prices. Many services are sold below cost today to win users. Promotions end (Gemini 3.8 Flash, for example, has a promotional price until the end of 2026). Plan as if costs could also rise.
3. Invest in skills, not bets. Knowing how to use these tools well will stay useful whatever happens to share prices. Picking the right stocks is a different job.
4. Paradoxically, a correction might help you. After the dot-com crash, bandwidth and servers were dirt cheap. If the AI bubble deflates, the over-built compute will make AI even cheaper for those who use it.

In one line

The bubble, if there is one, is in company valuations, not in the usefulness of the technology. People using AI to work better don’t need to predict the next crash: they just need to avoid depending on a single horse.

This is a personal opinion, not investment advice.

Sources

• Bloomberg Originals — Inside Anthropic, the $965 Billion AI Juggernaut
• Principles by Ray Dalio — Why Ray Dalio is Worried About an AI Bubble
• Bloomberg Podcasts — AI's Giant Money Loop Is Feeling Very 2008
• Bloomberg Television — AI's Next Big Risk Is Hiding in the Debt Market
• TestingCatalog — GPT-6 in ChatGPT (over 1.2 billion weekly users)
• Artificial Analysis — prices and cost per task

Informational content, current as of the date shown. Prices, models and rules change often: always check official sources. Written with the support of AI tools and reviewed by the author.