When a chief executive asks for trillions, not billions, when raising funds you know a sector may be getting a bit too hot.
In the long run, generative artificial intelligence will transform many industries and the way people work. But a report that OpenAI chief executive Sam Altman is talking to investors about an artificial intelligence chip project has raised a lot of questions.
A person familiar with the talks was cited as saying the project could require raising as much as $7tn. Scoring even a fraction of that figure — more than the combined gross domestic products of the UK and France — would seem a stretch, to put it mildly.
Nonetheless, it reflects just how hot the interest in AI, and the chips that power it, has become. The historical parallel that record-high AI-related stock valuations and fundraising targets bring to mind is the boom and bust in telecom stocks during the dotcom bubble era.
Back then, investors had expected the internet to transform the world. Telecoms companies and hardware suppliers would then be big winners. The problem was the sector’s valuations were pricing in that transformation to come almost overnight. Now, a similar level of optimism is driving investment in AI-related companies.
When the internet first became widely used, networking hardware was king. Servers needed to be built and connected using routers. Companies began building and buying hardware on the basis that extreme demand for servers would continue indefinitely. Telecom gear stocks such as Cisco surged more than 30-fold in the years to its 2000 peak.
But the collapse of the telecoms industry came earlier than expected — taking just four years to go from boom to bust — and much faster than the internet changed our lives. Oversupply pushed more than 20 telecom groups into bankruptcy by 2002. Shares plunged.
Now, in the world of AI, chips are king. Thus, the rush for AI companies to own more of the chipmaking supply chain is understandable. As AI models become larger, more chips are needed. A continuing shortage adds urgency.
Yet how long these shortages will last is debatable. It has been just two years since the world’s car industry was brought to almost a standstill because of a severe shortage of automotive chips. It took less than a year for that crunch to ease. Today, the supply of auto chips has not only normalised but many types are in a glut.
The biggest risk of throwing too much cash, too fast, at AI chips is overcapacity. That is already a problem for older-generation chips. With the current sector downturn lasting longer than expected, Samsung had to slash production last year to deal with a deepening chip glut. Japanese peer Kioxia posted a record $1.7bn loss for the three quarters to December. Adding to this, more than 70 new fabrication plants are being built.
Meanwhile, global silicon wafer shipments fell 14.3 per cent last year. Part of that is because of a cyclical downturn in the chip sector and a decline in demand for consumer electronics. But a slump in global chipmaking equipment billings, which fell more than a tenth in the third quarter, suggests future chip sector growth will remain at a more normalised level than what the AI boom has made us believe.
Another problem is that chips quickly become commoditised. Take, for example, the older 40nm chips used in home appliances. These are hardly in short supply today, but they too were scarce, cutting-edge resources when they were launched in 2008. As capital equipment depreciates, the price of older-generation chips falls.
Chips get faster and software more efficient every year. It took just two years for chips to upgrade from 7nm technology to the advanced 5nm used in the latest Nvidia chips. That rapid technological progress means companies may end up spending much less on chips in the future than they forecast today.
It is true there are clear differences between the dotcom era and the AI boom. For example, OpenAI’s revenues have already surpassed $2bn on an annualised basis, joining the ranks of tech’s fastest-growing platforms in history months after its launch. Today’s companies also have more ways to make profits.
But as with the early days of the internet, broader enterprise adoption of AI remains some way off. The transformation triggered by AI may take many years longer than today’s stock prices and funding expectations suggest. Hype and overinvestment are a dangerous combination. The way to avoid a similar fate to overhyped peers from the 1990s is to remember history repeats.
