What is the AI bubble, and what would a correction do to your tools?

What is the AI bubble? It's the fear that money poured into artificial intelligence, mostly data centres and chips, has run ahead of the profits AI earns. If investors lose patience, share prices could fall hard. This month the IMF warned markets are exposed to exactly that. For a small business, the question is what that does to your tools.

What worries people is the scale. Alphabet, Amazon, Microsoft and Meta are on course to spend more than $700 billion on capital expenditure in 2026, most of it on AI infrastructure. The same report said Amazon's trailing free cash flow had dropped 95% to $1.2 billion. OpenAI, the company behind ChatGPT, has committed to around $1.4 trillion of data centre spending over eight years, against about $13 billion of annual revenue, according to figures gathered on Wikipedia's AI bubble page. Somebody has to pay for all that eventually. Part of that somebody is you, through your subscriptions.

The short version

• The AI bubble is the fear that AI spending has run well ahead of AI profits, and in October 2026 the IMF warned that markets are exposed to a correction.
• Nobody can say when it'll burst, so plan on the assumption that a correction could land during any contract you sign this year.
• For small businesses the risks are price rises, suppliers closing, getting locked in to one supplier and contract terms you can't get out of. All four are already happening.
• You don't need to drop AI. Know what you pay for, test that your data exports, and keep your business logic in systems you control.
• A tool that can't show the hours it saves is the first one to cut when prices go up.

Will the AI bubble burst in 2026?

The warnings got louder in October 2026. The IMF's Global Financial Stability Report, published on 13 October, said markets are "increasingly exposed to an AI-driven correction". It pointed to circular financing, where AI firms act as each other's investors, lenders and customers, so trouble at one spreads quickly to the rest. A week earlier, Ray Dalio called AI a "classic bubble" and said of the moment it bursts: "I think we're close to that." And closer to home, the Bank of England's Financial Policy Committee warned in October 2025 that valuations "appear stretched, particularly for technology companies focused on artificial intelligence".

But the optimists have real evidence too. Google Cloud took $20.02 billion in the first quarter of 2026, up 63% on a year earlier, and Microsoft's AI business is running at more than $37 billion a year. That's money from paying customers, not projections. The dot-com crash didn't kill the internet. It killed the companies that couldn't pay their way, and the ones that could carried on trading.

How long will the AI bubble last?

Nobody can tell you the month, and even the IMF isn't predicting an imminent collapse. Dalio's argument is that rising interest rates and the debt behind AI spending are what eventually tip it over. Whether AI profits turn up in time to justify the spending matters just as much. Our view, as of 19 October 2026, is simple enough: don't try to time it. Plan as if a correction could land at some point during the contracts you sign this year.

What would an AI correction mean for the tools you pay for?

Price rises come first, and they've already started. Microsoft raised Microsoft 365 business prices on 1 July 2026, with Business Basic going from $6 to $7 a user a month and Business Standard from $12.50 to $14 at US list prices, and added Copilot Chat features to the plans. GitHub moved its premium AI requests to token-based charges from 1 June, so customers now pay for the amount of text the AI processes rather than a flat fee. Plenty of AI tools are priced below cost to win customers. Once investors care more about profit than growth, those prices go up.

The second risk is your supplier closing. Relay, an AI workflow automation tool pitched as the next Zapier, announced in July 2026 that it was shutting down. Free users lost access on 15 August and paying customers on 14 September. Why? OpenAI and Google had built similar automation into their own products. It isn't only start-ups, either. TechCrunch's running list of AI products that didn't make it includes OpenAI's ChatGPT Atlas browser, discontinued after less than a year, and quotes an S&P Global Market Intelligence estimate that about 42% of AI initiatives end up abandoned.

Then there's lock-in. Prompts, assistants and automations your team builds inside one supplier's platform usually can't be moved to another. If the price doubles or the service closes, you're rebuilding from memory.

Five checks to protect your business this week

None of this means pulling out of AI. The businesses that come through a correction comfortably will be the ones that know what they're paying for and how they'd leave. Most of it takes an afternoon.

1. List every AI tool you pay for. Include AI features buried inside software you already use, and anything paid for on a personal card. Watch for tools where only one person has the login.
2. Read the renewal terms. Look for auto-renewal dates, how much notice you get of a price change, and whether usage charges are capped. Be wary of annual commitments to small start-ups you couldn't afford to lose.
3. Test an export now. Download your data, prompts and workflow settings, then check you could actually use them somewhere else. Anything that only exists inside the supplier's app is a warning sign.
4. Keep your business logic somewhere you control. Your rules, templates and customer data should live in your own systems with the AI model plugged in, so changing supplier means changing a setting, not starting a project. Watch for agents that only run on one platform.
5. Put a number on what each tool saves. If a tool can't show the hours it saves, it's the first thing to cut when prices rise. Look hard at subscriptions nobody would miss.

Step four is where most small businesses are weakest, because it takes building, not just reading. At Swarm Labs it's the work we do most often in AI automation. We connect a model to your own systems through an API, the plug that lets two pieces of software talk to each other, so the model can be swapped out and your data stays where it is. Step five is why we built Time Hive. It logs every automation run and works out the hours saved, which turns a renewal decision into simple arithmetic.

Contracts deserve one more look. Ask what happens to your data if the supplier is bought or goes into administration, and get the answer in writing. Relay gave paying customers a few weeks' notice. Some suppliers give none, and an administrator's job is to look after creditors, not your data.

Keep the AI, lose the single point of failure

A correction would hurt AI investors far more than AI users, and the tools themselves aren't going away. What changes is the price, and which suppliers are still standing. If you can leave any supplier within a week and show that each tool earns its keep, a burst bubble becomes a renewal conversation rather than a crisis.

If you'd like a second pair of eyes on how much your business depends on one AI supplier, talk to us about it.

Sources

1. The Register: AI bubble could burst and take global economy with it, IMF warns
2. Yahoo Finance UK: Dalio says we're nearing point where AI bubble may burst
3. AJ Bell: AI firms vulnerable to sharp drop in valuations, Bank of England warns
4. Yahoo Finance: Hyperscalers hit $700 billion in 2026 AI spending plans
5. The Next Web: Alphabet and Amazon outpace Meta in AI as Big Tech's Q1 2026 earnings confirm cloud dominance
6. Wikipedia: AI bubble
7. TechCrunch: AI automation startup Relay shuts down, staff joins Google's Chrome team
8. TechCrunch: The AI graveyard, a running list of projects and startups that didn't make it
9. Office Watch: Microsoft 365 business price increase, who pays more in 2026
10. CIO Dive: What CIOs should know about AI-driven SaaS pricing changes