Forward Guidance Cold Turkey Comes With Withdrawal Symptoms - But The Patient (Investor) Always Recovers

It depends on what you mean by “AI bubble.” If the question is whether artificial intelligence itself is a passing fad, the answer is certainly no. Adoption is accelerating, investment is accelerating and the demand for the infrastructure behind it is real. 77% of companies use, or are exploring the use of AI in business. The bottlenecks, however, are real too.

If the question is whether all of the listed companies associated with AI deserve their current valuations, that is a different matter. Markets have a habit of taking a genuine long-term trend and pushing certain parts of it too far, too quickly.

The current boom is already substantial because it is capital-intensive. The hyperscalers are not simply spending on software licences, they are committing enormous sums to data centres, chips, power supply and physical infrastructure. Analysts expect that spending to increase further. J.P. Morgan Asset Management predicts the five major US hyperscalers will spend about US$700 billion on AI-related capital expenditure in 2026, with spending forecast to rise to around US$800 billion in 2027.

Global annual investment in data centers

IEA have tracked the rapid increase in worldwide datacentre growth, predicting this will reach US$900 billion by 2029.

IEA (2025), Global annual investment in data centres, Base Case, 2015-2030, IEA, Paris, Licence: CC BY 4.0

The better way to frame it is not that the market has reached a single, measurable bubble size. It is that expectations are becoming increasingly uneven. Some parts of the AI ecosystem may still be under-owned, particularly the less glamorous businesses helping to power and connect the build-out. Other parts are clearly being asked to deliver a great deal. That is where the real share valuation risk sits.

Do share bubbles always pop?

Not necessarily. Investors tend to imagine bubbles ending with one dramatic event: a violent sell-off, a recession, or a financial crisis that makes the excess obvious in retrospect. Sometimes that happens. But markets can also work off excess in less blatant ways.

A share price can go nowhere for years while earnings slowly catch up. Valuations can compress without a crash. Capital can move from the obvious winners into less visible parts of the market. None of that makes for a particularly satisfying headline, but it is often how markets adjust.

That’s important when talking about AI. The technology does not need to fail for some AI-related shares to disappoint. Equally, an expensive area of the market does not need to collapse for valuations to become more reasonable.

What history can teach us about stock-market bubbles

There is a tendency to look back at previous bubbles as though everyone should have seen the ending coming. Of course, once the share prices have collapsed and the failed companies have been identified, it all looks obvious.

In real time, it is more complicated. Railways were world-changing, even though railway speculation produced booms and busts. Computers were era-defining, although many early technology businesses failed. And the internet was transformative, despite the fact that the dot-com bubble created valuations that could not survive contact with reality.

The point is not that markets always get it right. They clearly don’t. The point is that markets can be wrong about the timing, the winners and the price, while still being broadly right about the direction of travel. That’s a useful way to think about AI.

Why the AI boom is different

The obvious comparison is with the dot-com era, when a genuinely transformative technology attracted vast amounts of capital and investors asked a great deal of some very highly rated businesses.

There are similarities, of course. But there is also an important difference. In the late 1990s, a great deal of the investment case rested on what the internet might one day become. It’s still achieving that potential decades later. With AI, adoption is already visible. Companies are using the tools, demand for computing capacity is rising and the people building the infrastructure are dealing with increasingly obvious constraints.

The bottlenecks are everywhere. Power is scarce in the right locations. Data-centre capacity takes time to build. Grid connections and digital infrastructure can take years. Environmental challenges must be solved. The supply of the most advanced chips remains tightly controlled. That does not remove valuation risk, but it does tell us that this is not simply a narrative being projected onto a few headline shares.

Watch earnings growth, not the calendar

Markets do not normally fall apart because investors decide that a chart has gone up for too long. They fall apart when the earnings that were meant to justify the price fail to arrive.

That is particularly important in an AI-driven market. Capital expenditures are rising sharply, and the return on that spending will eventually have to show up somewhere: in revenues, productivity gains, margins or some combination of the three. If earnings grow, the bubble needn’t burst.

Investors should therefore be wary of turning a valuation debate into a calendar forecast. Saying that a company trades at a high multiple of forward earnings is not the same as saying its share price must collapse next quarter. It may fall. It may also trade sideways while profits catch up.

The more important task is to watch the evidence. Are earnings estimates rising or falling? Are customers still spending? Are order books holding up? Those questions will tell us far more about the durability of the AI boom than any confident prediction about the date of a bubble burst.

Where the real risks lie

The most obvious danger is not the technology itself. It is the price investors are prepared to pay for exposure to it.

Markets are very good at taking a real theme and pushing the most obvious beneficiaries further than the fundamentals can comfortably support. That does not mean the theme is wrong. It means the margin for error becomes very small. A company can be doing everything broadly right and still fall sharply if the market had already assumed perfection.

There are other issues too; AI spending is placing heavy demands on corporate balance sheets, particularly where the investment is moving beyond internally funded capital expenditures and towards borrowing. At the same time, the equity market may have to digest more supply: fewer buybacks, more stock-based compensation and eventually some very large IPOs.

That matters because markets can become surprisingly fragile when liquidity is thin. The risk isn’t that AI will disappear. It’s that the market may become less tolerant of disappointment precisely when the capital needs of the boom are becoming more visible.

The future of AI: not a burst, but a recalibration

It’s tempting to try and predict the date when the AI boom will be confirmed or disproven. The problem is, it's rare for markets to provide that kind of clarity. Instead, the next phase is likely to be messier. Investors will become more selective about which companies can convert AI spending into revenue, profits and returns on capital. They will become less willing to reward vague promises. They may also begin to pay more attention to the less exciting businesses whose products are essential to the build-out.

That could mean difficult periods for some of the biggest names, greater dispersion within the S&P 500, or that the companies which performed best in the first phase of the boom are not necessarily the ones that perform best in the next.

But none of that requires artificial intelligence to be a bubble in the old-fashioned sense. Just as the technology is already reshaping business, it is also reshaping investment decisions across the economy. What changes from here is not the direction of travel, but the value of individual AI-related assets.

The boom is unlikely to burst. It is much more likely to become harder work.