July 2026 Performance Report

AI is booming. That doesn't make every AI stock a buy

Markets have had plenty of reasons to fall. Geopolitical shocks, stubborn inflation, higher bond yields and pockets of leverage are all capable of causing trouble. Yet the largest United States technology companies have kept producing the one thing that is difficult for investors to ignore: earnings.

There are signs of excess in artificial intelligence (AI), but there is also genuine demand, rapid revenue growth and substantial cash generation. Calling the whole sector a bubble is too simple. Buying anything with an AI connection is even worse.

A more useful analysis separates the businesses, examines who is funding whom and keeps watching the earnings.

One boom, several very different businesses

The AI trade is often discussed as though it were one investment. It is not.

At one end are the hyperscalers: companies such as Microsoft, Amazon and Alphabet. These businesses entered the boom with large established customer bases, strong balance sheets and enormous cash flows. Their cloud divisions are selling the computing capacity that companies need to build and run AI tools.

Their spending has become more aggressive. Businesses that investors once valued as capital-light software companies are pouring money into data centres, chips, power and associated infrastructure. In some cases, capital expenditure is absorbing most or all of the cash they generate.

That deserves scrutiny. But it is not automatically irrational. Revenue and profit from cloud and AI services have also been growing quickly. The uncomfortable truth is that a company can be spending an alarming amount of money and still be making a commercially sensible decision.

At another point in the chain are the suppliers. Nvidia and ASML have scarce technology and strong competitive positions. Memory-chip producers such as Micron, SK Hynix and Samsung are benefiting from powerful demand too, but their products are more exposed to commodity cycles. High current profits do not necessarily mean high-quality, durable earnings. These businesses have a history of over-investing when conditions are strong, creating too much capacity and driving the next downturn.

The distinction matters. Investors should not apply the same valuation framework to a platform with recurring revenue, a supplier with hard-to-replicate technology and a cyclical manufacturer enjoying unusually high margins.

The demand is real, but some of the financing is circular

The strongest case for the AI boom is straightforward: companies are paying for the technology because they expect it to save labour, improve output or create new products. Whether those expected savings are achieved will vary considerably between businesses.

If a business can replace a large recurring cost with a much smaller software bill, it may happily spend hundreds of thousands of dollars on AI services that did not exist in its budget two years ago. Multiply that decision across thousands of companies and the revenue opportunity becomes substantial.

The weaker part of the story sits among businesses that consume vast amounts of computing capacity while relying heavily on external capital. Some private AI developers can continue expanding while equity and debt markets are willing to provide funding. If that willingness fades, demand could slow quickly.

Circular financing makes the signal harder to read. When a chip supplier finances or guarantees a customer's data-centre capacity, and that customer uses the capacity to buy more of the supplier's chips, the order book no longer represents fully independent demand. The arrangement may work. It may even be necessary to build a new industry. But investors should recognise the similarity to vendor financing during the telecommunications boom: it looked self-reinforcing until capital markets turned.

The failure of one AI developer would not necessarily end the investment cycle. Its customers and computing capacity could move to a competitor. The greater risk is a broad loss of confidence that makes investors unwilling to fund the private companies at the centre of the ecosystem.

China may compress the economics

There is another reason to be selective. Chinese developers have released low-cost and open-source AI models that, in our assessment, are already capable enough for many tasks. They do not need to lead at the frontier to affect the economics of the industry. Even a model that trails the best systems but is available free or very cheaply can put pressure on prices.

China's manufacturing base may be even more consequential when AI moves into the physical world. Open-source vision and control software has reduced the cost and difficulty of building useful robots. Chinese manufacturers can combine that software with established supply chains and produce machines at price points that encourage experimentation by ordinary businesses, rather than only governments and large corporations.

This creates opportunity, but it may not accrue to the companies investors currently expect. Rapid adoption can coexist with falling unit prices and weak returns on capital. The technology can change the economy while disappointing shareholders in many of the companies that build it.

Higher bond yields change the calculation

AI investment does not occur in isolation. Data centres compete for capital, energy, water, construction materials and skilled labour. Heavy demand can create price pressure in particular inputs and regions, although productivity gains and falling computing costs may offset some of the broader inflationary effect. At the same time, large fiscal deficits and government debt issuance remain more direct influences on bond markets.

Higher long-term bond yields then raise the hurdle rate for every other investment. A future dollar of profit is worth less when discounted at a higher rate. Infrastructure, utilities and long-duration growth companies can all suffer, even when their underlying businesses remain sound. Investors may also demand more from equities when government bonds offer a better return.

Leverage makes sudden changes more dangerous. Momentum strategies, private credit structures and leveraged currency trades can look stable while prices move in the expected direction. When they reverse, forced selling can spread well beyond the asset that caused the initial problem.

None of this provides a reliable date for a market correction. It tells us where to look for fragility.

Why the decision is not all or nothing

The temptation in an uncertain market is to frame the choice as all or nothing. Either the AI boom is real and every related company must rise, or it is a bubble and cash is the only defensible position.

Both positions avoid the harder work.

There is always a convincing reason to stay out of markets, just as there is always a bullish story available somewhere. Waiting for every risk to disappear can leave an investor on the sidelines for years, while ignoring the risks can produce equally painful results. Diversification, deliberate position sizing and an explicit list of evidence that would challenge the thesis can reduce dependence on a single market-timing call. The appropriate exposure will differ between investors.

For the AI complex, earnings are central. Strong and broadening profits can justify continued investment despite high valuations and heavy spending. Deteriorating earnings, weaker returns on new capital, increasingly circular financing or a closure of private funding markets would be more serious warnings than dramatic headlines on their own.

The AI boom is real. So are the risks. Investors do not need to choose between blind enthusiasm and permanent pessimism. They need to distinguish durable cash flows from cyclical profits, independent demand from financed demand, and technological progress from shareholder returns.

That is less exciting than declaring a new era or predicting an imminent crash. It is also a better way to invest.