Is AI a bubble? What the evidence says, and what to do either way
Huge spending, rising debt and failed pilots on one side; real revenue and measured gains on the other. What it means for a small business owner.
Ali Hassan
· 7 min read
Quick answer
The investment side shows real bubble signs: record spending, growing debt, high valuations that the Bank of England has warned about, and many AI projects with no return. But the usefulness is also real: fast-growing AI revenue, wide adoption and measured productivity gains in structured work. For a small business, the answer is the same either way: start with one checkable task, keep commitments short, stay able to switch providers and measure the result.
A client asked me last week: "Everyone's saying AI is a bubble. Should I wait until it pops before I spend anything on it?"
It's a sensible question. AI stocks had a sharp fall in July, the spending numbers are enormous, and every week there's a new headline saying most AI projects fail. At the same time, people I work with are saving real hours with it every week.
So I went through the evidence from the people who actually measure this: company filings, central banks, and the research. Here's what it says, and what I think a small business should do either way.
The short answer: two different questions
"Is AI a bubble?" mixes up two questions:
- Is the money being poured into AI more than it will earn back, soon enough? That's about investors, share prices and data centres.
- Is AI useful for real work? That's about whether it saves your team time.
The answer to the first might be yes. The answer to the second is already yes, for some jobs. Both can be true at the same time. The railways and the early internet both had investment bubbles, and both turned out to be genuinely useful. A bubble bursting hurts investors. It doesn't make the useful parts disappear.
Signs of a bubble
- Alphabet spent $51.4bn on property and equipment in one quarter
- Bank of England: AI valuations still high, after a sharp fall in July 2026
- Analysts estimate $4.1 trillion of debt-financed AI spending from 2026 to 2030
- A widely quoted MIT report: 95% of companies saw no return from AI pilots
- Danish study: no measurable effect on earnings or hours yet
Signs of real value
- Google Cloud revenue up 103% in a year, to $30.7bn a quarter
- 88% of surveyed organisations now use AI in some form
- Generative AI reached 53% adoption in three years, faster than the PC or internet
- Support staff with an AI assistant resolved 14% more issues an hour
- Gains are biggest in structured, measurable work
Both can be true at once: the investment can be overheated while the tool is genuinely useful for the right jobs.
The case that it's a bubble
The spending is enormous
In its latest quarterly results, Alphabet (Google's parent company) reported spending $51.4 billion on property and equipment in a single quarter. That's more than double the same quarter a year earlier. Meta expects to spend $135 to $145 billion on capital expenditure in 2026 and says it will "continue investing aggressively" next year.
Stanford's 2026 AI Index puts global corporate AI investment at a record $581.69 billion in 2025, more than double the year before.
More of it is borrowed, and central banks are watching
The Bank of England's Financial Policy Committee noted in its September meeting that AI company share prices fell sharply in July, but that valuations remain high and rest on strong expectations of future earnings. It warned that a bigger disappointment around the pace of AI development or adoption could trigger a sharper fall.
The same record cites analyst estimates that about $4.1 trillion of AI-related spending between 2026 and 2030 will be financed with debt, and that AI-related debt issuance so far in 2026 was already more than double the whole of 2025. Borrowed money is what turns a fall in share prices into a wider problem.
Many projects aren't paying off
A widely quoted 2025 report from MIT's NANDA initiative found that 95% of organisations had got zero return from their generative AI efforts, as reported by The Register. It's worth knowing that this was based on 52 interviews, a survey of 153 people and a review of public announcements, so treat the exact number with care. But it matches what Gartner predicts: that over 40% of agentic AI projects will be cancelled by the end of 2027.
And when economists looked at the whole Danish workforce, they found no measurable effect on earnings or hours worked two years after ChatGPT launched.
The case that the value is real
People are paying for it
Spending only becomes a bubble if nobody pays for what it builds. In the same quarter as that $51.4 billion, Alphabet reported that Google Cloud revenue grew 103% to $30.7 billion. That's customers paying real money, and the release names sales of Google's own AI chips as a meaningful part of the growth.
Adoption is wide and fast
Stanford's AI Index reports that 88% of surveyed organisations used AI in 2025, and that generative AI reached 53% adoption within three years, faster than the personal computer or the internet.
It works, on the right jobs
The best-known field study followed thousands of customer support staff. Those with an AI assistant resolved 14% more issues per hour on average, and 34% more for newer staff. Stanford's summary of the research is that gains are largest in structured, measurable work where outputs are easy to check.
The flip side is that on complex, expert work, AI can slow people down. One careful study found experienced developers took 19% longer with AI, while believing they'd been faster.
Small businesses are still early
Here's the number that matters most for you. The US Census Bureau found that about 1 in 5 US businesses (19.8%) used AI as of May 2026, and that use among firms with fewer than 20 employees hadn't changed significantly over the previous months, while it grew in bigger firms. The hype is loud. Actual use in small businesses is still quiet.
What I'd do as a small business owner, either way
You don't need to predict the market. You need decisions that hold up whether the bubble pops or not.
- Start from a task, not from AI. Pick a job that's repetitive, frequent and easy to check. That's where the evidence says AI works. If plain automation does the job, use that. Here's the difference.
- Keep the commitment small. Monthly plans over long prepaid contracts, especially with young AI start-ups. If the funding dries up, some of them will close.
- Make it easy to switch models. Prices move. Google has already said one of its popular models will double in price from January 2027. Build so you can change providers without starting over. Here's what AI usage actually costs today.
- Own your data and your workflows. Run things on your own accounts, so nothing important lives only inside a vendor that might disappear.
- Measure it. If it saves time, you'll see it in the numbers within 90 days. If it doesn't, stop. I explain how in how to measure the ROI of automation and AI.
Not sure whether AI would help in your business, bubble or not? Book a free 20-minute call. Tell me where your team's time goes, and I'll tell you honestly where AI would pay off, where plain automation is enough, and where I'd leave things as they are.