At the height of the Dutch tulip craze in 1636, a single tulip bulb was bought for 2,500 florins – or, as the bill stated, eight fat swine, 12 fat sheep, four tuns of beer, 1,000lb of cheese, a complete bed, a suit of clothes and a silver drinking cup, plus the change.

Next generation: NTT’s Dagenham facility is among the latest wave of data centres under development in the UK

A year later, the price bubble burst, with tulip bulbs becoming almost worthless.

On the surface, ChatGPT and its ilk have little in common with tulips – until you compare today’s burgeoning artificial intelligence (AI) industry with the market dynamics of the world’s first documented financial bubble.

Last week, the Bank of England (BoE) cautioned against a ‘sharp market correction’ due to a potential financial bubble involving AI investment and over-inflated asset values for leading tech and AI firms.

The BoE’s financial policy committee report, published on 8 October, notes: “On a number of measures, equity market valuations appear stretched, particularly for technology companies focused on AI. This, combined with increasing concentration in market indices, leaves equity markets particularly exposed should expectations around the impact of AI become less optimistic.”

In essence, a bubble occurs when the asset value of a company (or commodity) far outweighs the value in terms of annual revenue or fundamental expected returns.

There will be winners and losers and market corrections undoubtedly
Andrew Fray, Cushman & Wakefield

The investment figures around AI and data centres are certainly mammoth. Earlier this year, Morgan Stanley estimated that £2.2trn would be spent on data centres that support AI worldwide up to 2029.

AI company OpenAI’s investment alone is estimated to have topped $1trn (£751bn). But in August, OpenAI chief executive Sam Altman warned of an emerging bubble, saying that “smart people” had become “over-excited”. OpenAI itself was valued at $500bn (£376bn) in August, but posted revenues of only $20bn (£15bn). The firm is yet to turn a profit.

A number of big deals in the sector have involved tech companies investing in other tech firms that are their customers. For example, on 22 September, tech giant NVIDIA announced it would invest $100bn (£75bn) into OpenAI to support data centre and power capacity – and with it, demand for its own chips. With tech firms pumping money into their own sector in this way, valuations can become inflated and increasingly unclear.

Implications of an AI crash

Many are now asking what will happen if the possible AI bubble pops. Analysts look to the dot-com bubble – the bursting of which led to a recession and a spike in US office vacancy rates – as the main example for the implications of an AI crash.

Andrew Fray, head of the EMEA data centre advisory team at Cushman & Wakefield, notes that the dot-com bubble and subsequent crash were largely a progression from the telecoms era, pointing out the similarities between tech firm Cisco then and NVIDIA now.

“[Before the dot-com era] it was all about voice traffic, then when the internet came in, the thing that accelerated the adoption of the internet was the Cisco router, which could control data traffic, and Cisco became one of the most valuable companies on the planet.

Search for revenue: OpenAI has invested over £1trn but has yet to make a profit

“With NVIDIA, you have this chip that controls AI – and it is disruptive. There will be winners and losers, and there will be market corrections, undoubtedly.”

Fray adds: “There will be different innovations and business cases, just as for the internet. The difference is that the speed of innovation and change possible with AI is going to be breathtakingly quick.”

However, many in the market believe the UK’s data centre sector looks to be resilient to any possible tech downturn – mainly because the AI boom has yet to really take off on UK shores.

Rupert Duckworth, associate director in Savills’ EMEA data centre team, says: “Up until now, the majority of demand for data centres [in the UK] has been driven by public cloud [computing] services such as AWS [Amazon Web Services], Microsoft and Google. That’s the most profitable part of their business and has had steady growth since its inception in 2012-13.

AI is a completely different beast – the difference is the location
Rupert Duckworth, Savills

“AI is a completely different beast – the difference is the location. We are at the infancy of this in Europe. A number of sites have been purchased, but very few have actually started construction. The requirements [for AI centres] are across the whole region [of Europe], whereas public cloud data centres are very specific to geographical locations.”

Fray echoes this view, saying the UK has “a very well-established, traditional data centre estate”, while “London in particular – along with Amsterdam, Paris and Frankfurt – has led the way in terms of cloud adoption”.

Significant headwinds

Fray adds that when it comes to developing the next generation of AI data centres, “we have significant headwinds in terms of constraints about where you can find more land and power to go to the next level”.

The challenges of executing vast data centre expansion plans are becoming apparent in both the UK and the US. The limitations of power generation are well documented; Goldman Sachs has estimated that an extra 300,000 jobs are needed to meet US power demands – mostly coming from data centres – by 2030.

Lawrence Turner, director at planning consultancy Boyer, says that in the UK, the current tech investment cycle has moved so fast that all the action has occurred within half the timeline of a typical Local Plan.

He also believes the UK is well placed, with a shortage of facilities in most areas rather than a glut of empty data centres littering the country. He adds: “Say the bubble bursts: do we have an over-allocation of land for data centres in the UK? Do we have a whole construction industry geared up to deliver supporting services and supporting industries around data centres?” His answer to both these key questions is ‘no’.

Another possible constraint on AI sector growth could be the challenge of generating enough revenue to meet its large cashflow requirements. In August, Harris Kupperman, founder of US hedge fund Praetorian Capital, calculated in a blog that the AI industry needed around $480bn (£360bn) of revenue to hit its target return in terms of 2025 capital expenditure alone.

The problem is whether there are actually enough customers on the planet to match that income ambition.

Meanwhile, in the UK, the tech sector has supported the upcoming AI bill, which is intended to help establish the investment pathway for AI and data centres.

Duckworth says: “Regulation is going to be very important for AI. In the US, there’s basically no regulation, while in Europe it’s very stringent. From the rumblings we’re hearing, [the AI bill] is likely to fall somewhere between what Europe and the US have, putting the UK at a massive competitive advantage [over Europe].”

But whether the sector’s growth is sustainable will depend on factors outside the government’s direct control – such as the possibility that the US stock market, currently buoyed by tech stocks, will suffer a massive crash when the AI bubble bursts.

According to Fortune magazine, AI-related firms account for 75% of US stock market gains so far this year, while by some estimates, the so-called ‘magnificent seven’ US tech firms – Apple, Amazon, Alphabet, Meta, Microsoft, NVIDIA and Tesla – make up over a third of the US S&P 500 stock index.

In the 1630s, Holland’s tulip traders eventually ran out of buyers able to support the market, resulting in its collapse. Many will be hoping the AI industry doesn’t outpace its own market and suffer a similar fate.