AI bubble set to inflate further

There are several truisms when it comes to tech bubbles. One is that it’s often difficult to see one when you are inside it: each individual spending or investment decision can seem rational, even if the net effect looks extreme.

When there’s general agreement that a bubble is forming, it often has much further to inflate and investors who bail early miss out. And after a bubble has burst, it can take years to work out whether it was just the product of hype or the precursor to an even bigger tech boom ahead.

As US tech companies enter their latest earnings season, a distinct giddiness has started to creep into tech valuations on Wall Street. The middle of 2024 was always destined to be something of an “air pocket” for generative artificial intelligence. The investment boom triggered by the technology has been all too visible, but it takes time for all that new capacity to be put to productive use by the tech industry’s end customers. Wall Street’s patience is about to be put to the test.

Software companies, which are best positioned to capture the value of the new technology by embedding it as features in their existing products, are struggling under a stock market cloud.

Nor has AI led to any compelling new services for consumers, even if Apple recently promised to sprinkle the technology liberally across its devices. For many people, discovering ChatGPT brought a frisson of interest. But unlike the first time they picked up an iPhone, used Google’s search box or found their friends on Facebook, it hasn’t changed their digital lives.

At best, this points to a lag in the widespread take-up of generative AI. At worst, it shows that the technology isn’t as transformative as tech companies have claimed. The longer the pause, the more glaring the gulf between investment boom and lacklustre end demand will become.

Accentuating this is the narrowness of the base on which this investment boom has been mounted. Nvidia said late last year that cloud companies — a market dominated by a handful of big players — accounted for more than half its sales of data centre chips. Any hint from the big tech companies during this earnings season that they are moderating their spending would deal a serious blow.

Yet as tech companies prepare to announce their latest earnings, all the signs are that the boom is still in full swing. Many business customers have barely begun their first pilot projects using the technology and will be increasing their testing of the technology in the coming months, even if it’s unclear what ultimate uses they will find for it. Pouring money into large language models and the infrastructure to support them has also become a strategic necessity for the big tech companies themselves. If machines that can “understand” language and images represent an entirely new computing platform, as many in the tech world expect, then all the big players will need a stronger foothold in the technology.

It’s also worth noting that these companies have more than enough financial firepower to maintain and even escalate the battle. The combined operating cash flow of Apple, Microsoft, Alphabet, Amazon and Meta jumped 99 per cent in the past five years, reaching $456bn in 2023. That was more than enough to accommodate capital spending that ballooned by 96 per cent to $151bn.

Meanwhile, the next big product cycle for chipmaker Nvidia, based on its new Blackwell chip architecture, isn’t even due to begin until the second half of this year. The lower costs this promises for training and running large AI models have guaranteed a stampede from customers, even as demand for its earlier generation of chips remains strong.

There’s a paradox here common to all new tech platforms. As the cost of using the new technology plunges, customers could theoretically get away with buying less. But rapidly falling costs usually lead to new uses being found, and demand instead soars. As with everything to do with generative AI, this is happening at warp speed, and it is hard to tell how closely this will mirror other disruptive tech cycles.

At some point, of course, all of this investment has to earn a return. If not, then the chief executives who have been pushed by their boards and by fear of competition to demand their companies find uses for generative AI will eventually lose patience and move on. But all indications are that we’re not at that point yet.

richard.waters@ft.com