You are carrying bets you never placed

Do you know how much of your business now depends on someone else being right about AI?

Not your AI plans. Theirs. The software you pay for every month, the supplier whose quote assumes it can do the same work with half the staff, the customer whose budget for you was approved on the back of their own AI roadmap.

In February 2026, Yahoo Finance reported that Microsoft, Alphabet, Amazon and Meta expected to spend between US$635 billion and US$665 billion on capital expenditure this year, driven by AI. A Bain & Company report in September 2025 estimated the industry needs $2 trillion in annual revenue by 2030 to fund the computing it expects to need — and even counting savings, would still be around $800 billion short.

This is not a market call. Nobody knows whether that gap closes. Your business should not need to know.

Almost every owner I meet has an opinion on whether AI is a bubble. Very few have checked whether they are standing in it.

Ask the owner's question, not the investor's

Investors ask whether prices are too high. Owners need a narrower question. Which of your costs, contracts, suppliers and customers are now tied to someone else's bet.

When a boom unwinds, the damage is not spread evenly. It lands on the businesses that borrowed against the boom — that signed the long commitment, hired ahead of the revenue, or built their margin on a price that only held while money was easy.

The businesses that simply used the technology are inconvenienced. The ones that borrowed against it are exposed.

Using the technology is a decision. Borrowing against it is a bet.

So stress-test your own exposure. Not the market's. Yours. Run three scenarios through your numbers.

• Your AI tool prices double. List every subscription and usage bill, double it, and see what is left of your margin. Today's price is a decision someone else made, and they can make it again.
• A key vendor folds. Pick the one tool your team could not work without for a week. Find out where your data sits, how you would get it out, and what you would run instead.
• Your biggest customer cuts spend by a third. If their budget rides on their own AI program, yours does too. Work out how many months you could carry the gap.

Then fix whatever breaks first. Usually it is a contract term or a cash buffer thinner than it looks.

None of this says the technology is not real. It is. The internet was real in 2000 as well, and it changed nearly everything — after a crash that took plenty of sound businesses with it.

A bubble in prices is not a verdict on the technology. It is a verdict on the timing.

Being right about the technology is common. Surviving the timing is not.

Amazon raised the money before it needed it

In February 2000, Amazon sold $672 million in convertible bonds to investors in Europe. Brad Stone tells the story in The Everything Store (2013). The chief financial officer, Warren Jenson, wanted more cash on hand because he feared nervous suppliers might demand faster payment if the company ever looked shaky.

It was not cheap money. The bonds carried a higher interest rate than Amazon's previous raise, and the market was already starting to wobble.

The Nasdaq peaked a few weeks later, on 10 March 2000. By Stone's account, three weeks later the deal would not have gone through. Amazon's shares went on to lose more than ninety percent of their value over two years, as CNBC later recounted.

By June 2000, a Lehman Brothers analyst was arguing the company would run out of cash within a year. It did not. The cushion was already in the bank.

Read that as Amazon being right about the internet and you have misread it. Almost everyone in 2000 was right about the internet. Plenty of them did not see 2002.

Amazon survived because it secured cash while the window was still open. Jenson was not forecasting the crash. He was asking what happened if the people Amazon depended on got nervous — and paying up so the answer would not matter.

Why I ask it first

I run this test on every business I work with, and I have since the hardest stretch of my career.

I spent six years leading a development company backed by a private family office. Then came the COVID lockdowns, and after them material and labour costs went through the roof while borrowing costs kept mounting.

It fell apart. I was out of a job, with no salary, a young family and a pile of bills.

What it taught me is plain. A development is exposed to a cycle nobody inside the business controls — construction prices, interest rates, the appetite of lenders. You can run the work well and still be flattened by the numbers around it.

So I carry one question into anything I sign: what happens to us if the cycle we rely on turns. I ask it of every cost line, contract and customer before I ask whether the opportunity is any good.

The opportunity usually is. The exposure is where businesses get hurt.

Do this before Friday

Open last month's bank statement. Circle every payment to an AI tool, and every customer whose spend with you depends on their own AI plans.

Then run one scenario only: your biggest customer cutting spend by a third. Work out, on one page, how many months you could carry it.

If you do not like the number, you have found the first thing to fix.