![](https://dl6pgk4f88hky.cloudfront.net/2024/08/28/GettyImages-2089838868-1038x778.jpg)
On Wednesday 28 August the processor and software company Nvidia announced that in the second quarter of this year, its revenues had risen steeply – up by 15 per cent in three months, more than doubling in a year – to ten billion dollars a month. Share buybacks and dividends were returning billions to investors in a company that currently appears to have an unassailable position in the perhaps the world’s fastest-growing technology sector – generative AI.
What happened next defied the logic of the market. A stellar earnings report is supposed to lend wings to a company’s stock, but Nvidia’s share price fell sharply (by 6.4 per cent) the day after its earnings were released, and they have continued to fall. Yesterday (Tuesday 3 September) it fell further still; Nvidia’s CEO, Jensen Huang, is estimated to have lost $10bn in a single day.
The main reason for this apparently backwards situation is that the AI revolution has been priced in, even though it hasn’t actually happened. Nvidia represents a huge gamble by investors around the world, not only on a single technology – generative AI – but on a single company. It is the leading provider of the hardware and software being used to create the large language models that Big Tech has proclaimed to be the next big opportunity.
Nvidia’s explosion in value was prompted by the release of a single piece of consumer software: ChatGPT. As the chatbot became a global news story, companies scrambled to avoid being left behind in the new technological revolution, and Big Tech began pouring money into hardware – particularly Nvidia’s H100 chips – to develop their own models. In 2023, Meta and Microsoft each bought around 150,000 of these chips, which typically cost around $30,000 each. Google, which designs its own AI chips, is nonetheless one of Nvidia’s biggest customers.
Nvidia’s value to investors soared in response: at the beginning of this year it reached a market capitalisation of £2.56 trillion, which is greater than the entire UK economy and five times the market value of JP Morgan, the world’s largest bank.
This was partly a matter of luck. Nvidia has been making GPUs (graphics chips) and associated software since the 1990s; for decades, PC gamers have upgraded their machines with its GeForce graphics cards. The central processing units that are the brains of most computers calculate in sequence, but graphics processing involves doing lots of simpler calculations at the same time – known as parallel computing – and GPUs excel at this work. The artificial neural networks that computer scientists began building in the early 2010s process information in a similar way, and Nvidia has increasingly specialised in creating the extremely complicated hardware and software on which these models can be built. In the process it has established a leading position in the industry many see as the biggest technological advance since the internet (or the wheel, or fire, depending on how evangelical your faith in the AI bubble is).
This is the kind of “economic moat” of which investors dream, and the market confidence in Nvidia has been historic. Every £100 invested in the company shares in the week that OpenAI made ChatGPT available to the public in November 2022 would have returned a profit of around £700 if sold last week. The release of the company’s 2023 earnings, in February 2024, caused the largest single-session increase in the value of any company on record, adding $227bn of market value in a single day.
However, Nvidia’s success has also meant that investment in financial markets – which was already heavily focused on Big Tech – has concentrated more than at any time since the dotcom crash. Nvidia’s customers are Meta, Microsoft, Google and Amazon, so the exploding revenues of the market’s most important stock are being paid mostly by the other companies at the very top of the market.
The central issue in this situation is that tech valuations have been kept aloft not because Microsoft and Alphabet are currently making vast profits from AI, but because they expect to. Tens of trillions of dollars are focused on the highly related fortunes of a handful of companies in an industry that has not yet proved itself profitable. Financial markets have become a roulette game in which almost all the chips are bet on one square.
With so much riding on a single lofty expectation – the unproven and very possibly false idea that a specific type of AI will perform a productivity miracle – any item of bad news, such as the antitrust investigation now being conducted by the US Department of Justice into Nvidia’s dominance of AI chipmaking, can tip the needle from greed to fear in an instant.
Plenty of analysts expect this to end in a similar manner to the last technology bubble. Diana Iovanel, senior markets economist at Capital Economics, told me her team’s base scenario is that the “AI-fuelled equity bubble” will continue to inflate as Nvidia and others reap the rewards of investor confidence, and that by late next year, valuations will reach earnings multiples last seen “during the peak of the dotcom bubble”. “Eventually,” Iovanel told me, “we expect this bubble to burst, with expected earnings and especially valuations falling significantly.”
Perhaps, as a normal person, you’re wondering why you should care. The answer is that you almost certainly invest in Nvidia and its Big Tech colleagues: about 10 per cent of the typical UK pension is invested in the “Magnificent Seven” tech stocks, according to PensionBee. There’s a good chance your pension has more exposure to Nvidia than any UK company, and when the AI bubble pops, the value of your retirement savings could reflect that.
In the dotcom crash, professionals saw the irrational exuberance (a phrase coined to describe the frothy valuations of the time) for what it was, and quietly sold out while regular investors – through their passive retirement savings – were left holding shares in soon-to-be worthless companies. Some variation of this pattern may be about to happen again.
[See also: When the AI bubble bursts]