Picture credit: Wong Yu Liang/Getty

What if the AI bubble bursts?

Arguing that an AI bubble is a good thing reeks of techno-optimist complacency

Artillery Row

“If people had told me this was gonna happen, I was like, ‘Man, you must be smoking some really good crack.’” These sage words were the reaction of Elon Musk to having become the world’s first trillionaire. 

The spectacular success of the initial public offering of Musk’s SpaceX — based partly on the aerospace giant’s acquisition of xAI and his pledge to put AI data centres in space — arguably marked the latest peak of an AI-driven economic boom which is looking increasingly detached from fundamentals. A June survey by Bank of America found that the share of fund managers who considered an “AI bubble” to be the biggest risk facing markets had increased by 400 per cent since May, and ever more commentators are warning that bullish growth may already have tipped over into irrational exuberance. 

But some economists have argued that an AI bubble might actually be a good thing, even if it’s doomed to burst. After all, the huge number of AI data centres currently being built are bound to be useful in the long term, even if the investors building them are soon wiped out. 

Join Britain’s most civilised publication.

Challenge the consensus. Access rigorous analysis.

Archive article

Don't worry. You can continue reading by subscribing to get full access.

Subscribe

Already a member? Log in.

Premium article

Don't worry. You can continue reading by subscribing to get full access.

Subscribe

Already a member? Log in.

Subscribe Now

Those who advance this view are overlooking the crucial role of the hidden messages communicated by profits and losses, booms and busts

If this were true, it would make the current AI boom an example of a so-called “productive bubble”, but arguably not the first such example. In the 1840s, Britain was gripped by the infamous Railway Mania, which saw Parliament authorise 263 new railway companies and nearly 10,000 miles of new track in the year 1846 alone. This white-hot bubble inevitably burst in the calamitous Panic of 1847, but once the dust had settled Britain was left with a glittering new rail network which was put to use in the ongoing Industrial Revolution. More recently, the Dot-Com Bubble of the 1990s culminated in the 2001 recession and arguably sparked the disastrous housing bubble of the 2000s, but at least it also resulted in tens of millions of miles of useful fibre-optic cable being laid. 

It would be comforting to believe that even the most delirious of speculative bubbles may ultimately benefit the economy. However, those who advance this view are overlooking the crucial role of the hidden messages communicated by profits and losses, booms and busts.

The bursting of an economic bubble is simply the collective manifestation of many businesses simultaneously turning out to be unprofitable. But in a market economy, whether or not a business is profitable is a meaningful signal, not just a matter of random luck. Consumers are willing to pay higher prices for goods which they value more highly or need more urgently. This enables the producers of highly-valued goods to pay more than the producers of less highly-valued goods for the inputs (labour, material ingredients, etc.) needed to produce those goods, bidding up the prices of those inputs. If a producer of less highly-valued goods can’t convince consumers to pay enough for its goods to enable it to afford those high input prices, that producer is, by definition, unprofitable.

In other words, if a business is profitable in a market economy, this indicates that it is using scarce resources in a way which caters to the most urgently-felt needs of consumers (and we are all consumers). But if a business is unprofitable, this indicates that it is wasting resources from the collective perspective of society. Importantly, just because a business is unprofitable, that doesn’t necessarily mean the goods it produces are totally useless. But if it can’t sell its products for more than the price of the resources required to produce them, this is a signal that those resources are more urgently needed elsewhere.

One clear example of useful goods being produced in a clearly wasteful way can be found in China’s infamous “ghost cities”. Following the 2007-08 global financial crisis, the Chinese government initiated a surge of credit expansion which created an illusion of abundance and drove investors into a building frenzy which far exceeded demand. By 2017, an estimated 65 million homes in China lay vacant, around 20 per cent of the total urban housing stock of the country, with many of these properties remaining empty for years afterwards.

The supposedly “productive” dot-com bubble of the 1990s was also stimulated by a period of credit expansion, which distorted perceptions of demand and expectations of profitability. After the crash, it was estimated that as much as 95 per cent of the fibre-optic cabling which had been laid was unused “dark fibre”. It’s unclear how much dark fibre remains beneath our feet today, but as late as 2007 around 66 per cent of the cabling installed during the dot-com bubble still lay unused. 

One sense in which the believers in a “productive bubble” are correct is that the data centres currently being built will probably still be used after the bubble bursts. But the relevant question is not whether people will make the best of what they already have once the money has been spent. The relevant question is whether consumers’ most urgently felt desires would have been better satisfied by spending the scarce resources of the economy to produce data centres of a different kind, in a different location, or at a different time, or perhaps to produce something else entirely. 

In economics, the concept of “path dependence” describes the ways in which economic and technological progress, once it has begun on a less-than-ideal course, usually ends up continuing in that direction, stuck in a rut thanks to the way past decisions limit the options available in the present. If we are indeed in a bubble and AI infrastructure is being produced in excess of demand, the instinct to make the most of that existing infrastructure once the bubble has burst will only disincentivise the development of different and potentially superior technologies. In this way, the supposed silver-lining of data centre capacity may actually be the source of the most lasting damage to emerge from the current AI boom, limiting growth long after the bubble bursts.

Archive article

Don't worry. You can continue reading by subscribing to get full access.

Subscribe

Already a member? Log in.

Premium article

Don't worry. You can continue reading by subscribing to get full access.

Subscribe

Already a member? Log in.

Enjoying The Critic online? It's even better in print

Subscribe today to Britain's most civilised magazine

Subscribe
Critic magazine cover