Is AI a bubble?
Five tech giants are expected to invest about 800 billion dollars this year, and AI takes most venture money. What bulls and bears say, and what to watch.
The question comes up in every boardroom: is AI a bubble? Even the people at the centre ask it. In August 2025, OpenAI's chief executive Sam Altman told reporters: “Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes.” He also gave the best short description of how it happens: “When bubbles happen, smart people get overexcited about a kernel of truth.”
A bubble, in plain terms, is when prices and spending run far ahead of the cash a thing will ever produce. Few doubt that the technology is real. The open question is whether the money now being spent on it can earn a return, and how soon. Here are the numbers, the arguments on both sides, and what we watch.
The scale of the spending
Start with the buildings. J.P. Morgan Asset Management puts the capital spending of Alphabet, Amazon, Meta, Microsoft and Oracle at 154 billion dollars in 2023, 241 billion in 2024 and 416 billion in 2025. Analysts' consensus expects 799 billion in 2026 and 1'093 billion in 2027. These are total investments, not only AI, but the jump coincides with the race to build data centres for it.
Single companies show what that means. Microsoft spent 41 billion dollars in one quarter, and its finance chief, Amy Hood, said roughly two thirds of it went to short lived assets, “primarily CPUs and GPUs”. Oracle, which has signed large AI contracts, reported negative free cash flow of 23.7 billion dollars for its year to May 2026, and raised 43 billion dollars of debt in that year.
Venture capital followed. Crunchbase counted 425 billion dollars of venture funding in 2025, roughly half of it for AI, including the largest private round of all time: 40 billion dollars for OpenAI. In the first three quarters of 2026 the total reached 679 billion, and in the third quarter AI took 64 percent. In May, Anthropic raised 65 billion dollars at a valuation of 965 billion.
Money that goes round
What worries even optimists is how much of the money moves in circles. In September 2025, Nvidia and OpenAI signed a letter of intent under which Nvidia “intends to invest up to $100 billion in OpenAI as the new NVIDIA systems are deployed”. The supplier funds the customer, and the customer buys the supplier's systems. Two weeks later AMD gave OpenAI a warrant for up to 160 million AMD shares, vesting as OpenAI deploys AMD chips and as AMD's share price reaches certain targets.
Deals like these are public and disclosed. In our view, though, they can make demand look more independent than it is, and they tie the fortunes of the largest companies more closely together. In September 2026 the Bank of England's Financial Policy Committee pointed to “the increasing indebtedness of AI firms combined with opacity and, at times, ‘circular arrangements’” as something that makes the risks harder to judge.
What the bears say
The bears start with prices and debt. The same Committee noted that equity valuations for AI companies fell sharply in July, cited a Morgan Stanley estimate that AI related debt issued worldwide had reached around 450 billion dollars by early September, more than double all of 2025, and an estimate by J.P. Morgan analysts that debt will finance around 4.1 trillion dollars of AI investment between 2026 and 2030. Its conclusion: “The risk of a sharper correction persists.”
They add concentration. At the end of September, the ten largest companies made up 40.6 percent of the value of the S&P 500 but earned 36.1 percent of its profits, and industries linked to AI accounted for 53.2 percent of the index, J.P. Morgan calculates. A correction in one theme would reach most portfolios that track the index.
And they point to the life of the assets. A data centre lasts decades, but the chips inside it age within a few years. If they lose their value faster than the accounts assume, today's profits are overstated and the bill arrives later.
What the bulls say
The bulls answer with demand that can be measured. Microsoft's Azure cloud grew 43 percent in the quarter to June, and Amy Hood said: “Customer demand continues to exceed available capacity.” Anthropic said in May that its run rate revenue had crossed 47 billion dollars. Much of the capacity is sold before it is built: Oracle reported 638 billion dollars of contracted revenue not yet earned, most of the recent increase from large AI contracts in which the customer paid in advance for the chips or supplied them.
They also point to valuations. On J.P. Morgan's figures, the ten largest companies in the S&P 500 trade at 20.9 times expected earnings, against an average of 20.8 since 1996. By that measure, the largest companies are priced in line with their own history, and together they earn more than a third of all the profits in the index.
Our reading, and what to watch
Both sides describe the same thing. The technology is real, the demand is measurable, and parts of the financing look like a bubble. Railways and the early internet built infrastructure that lasted, while many of the investors who paid for it lost money. AI may well do both at once.
We watch four things: the free cash flow of the builders, how much of the spending is financed with debt and private credit, how much revenue comes from customers who are also investors, and whether AI revenue keeps pace with consensus spending of over a trillion dollars in 2027.
For a company that uses AI rather than builds it, the useful question is smaller. Is your own AI spending earning its keep? That is what North is for: every number that matters in one view, from the first visit to the last invoice, each one measured or marked as missing. Whether the market is in a bubble is for investors to argue about. Whether your AI pays for itself is a number you can know.
