Opinion: AI slowdown calls may be justified while bubble collapse could loom

Warnings about powerful and uncontrolled AI have dominated the headlines this past week. But even amid the omens of apocalypse, it is still worth asking if the entire AI-industrial complex is sailing towards a financial iceberg.

It will hardly be our most pressing concern if the bots are poised to take over the world, but a collapse of the AI bubble would have repercussions far beyond the US.

AI bosses have united in recent days to warn us the immensely powerful product they have sunk billions into creating may or may not destroy the human race.

Some are also calling for limits on how rapidly the technologies should be allowed to progress, and how they can be used.

There is ample evidence that AI urgently needs regulating – from the ability of Meta’s pervert glasses (sorry, smart glasses) to film us without consent to the lack of adequate safeguards that allowed swarms of chatbots to go on a hacking spree.

Some of the suggestions currently being mooted, including independent analysis of AI models, seem important improvements on the ungoverned status quo.

But we should also be alert to the risk that a small number of intricately linked megafirms that have racked up multibillion-dollar debts are hoping the state will throw a regulatory moat around them.

Perhaps a government-backed AI “pause” could prevent cheaper Chinese options from encroaching on Silicon Valley’s market dominance, for example.

Away from the risk that their product might obliterate human life, there are more old-fashioned reasons for tech bosses to be fearful – including the fundamental one of whether the economics of their businesses work.

The first worry is the sheer scale of debt issuance being used to fund the hectic pace of the datacentre rollout by the hyperscalers building them – Google, Amazon, Microsoft, Meta and Oracle – $132bn (£99bn) this year alone on one estimate.

In a world of fragile bond markets with yields on 10-year US treasuries, a global benchmark for borrowing costs, hovering at about 5%, the size of these debt piles could be one potential trigger for a market rethink.

That is especially true given the second reason for concern: the fact that the “unit economics” of AI continue to be questionable and aren’t moving in the right direction.

As a recent Bloomberg report put it: “The price of AI is collapsing, while the cost of building it is not.” It pointed out that OpenAI has repeatedly cut its fees to hang on to customers, for example.

An index by the research company Silicon Data that aims to track how much customers are paying for a million tokens – the units of data processed by large language models – shows it more than halving since June, to less than $1.

Yet at the same time, frenzied demand for the real-world components of datacentres – semiconductors, for example – is keeping costs elevated.

The maths only works, it seems, on the assumption of epic revenue growth. Anthropic apparently told investors recently that its “adjusted operating income” was positive – the only problem being that this measure effectively excludes many of its costs.

As the digital rights campaigner Cory Doctorow puts it: “These companies are claiming that they are so cool that their profitability can only be measured using a novel, secret form of mathematics.”

Some aspects of this edifice may seem familiar to veterans of the global financial crisis: masters of the universe, with a business model that is so clever mere mortals can’t be expected to understand it, all propped up with ample dollops of leverage.

A frankly terrifying recent research note from the financial analyst Groundbreaker made this analogy clear, by unearthing a third reason to worry – just totting up debt underrepresents the financial promises underpinning the boom.

It sets out the $1.5tn “compute commencement wall” faced by the AI labs over the next couple of years, drawing a comparison with the moment when cut-price “teaser” mortgage rates began to run out in 2007 and 2008.

When those rates ended, flipping low income homeowners on to much higher rates, borrowers began defaulting in their droves, lighting the touchpaper on what became the global financial crisis.

According to Groundbreaker’s analysis, in many cases datacentres are being built and kitted out on “take or pay” contracts, with not a dollar due until a deadline is hit (often two to three years), and they hum into life.

In the interim, the hyperscaler building the datacentre books the value of the contract as expected future revenue, and shareholders cheer. Meanwhile, the buyer, (a frontier lab such as OpenAI or Anthropic) doesn’t yet have to account for the costs it will have to pay, once it starts using it.

The analysis suggests the abrupt jump in costs, as contracts mature and datacentres come online, could be an eye-watering $700bn next year and more than $800bn in 2027.

That could all be just fine if revenue continues to rocket. Or perhaps not so much, if AI’s end users aren’t prepared to pay enough to cover the costs – perhaps because cheaper options arise.

This isn’t, technically, debt – but the impact, if obligations can’t be met in full, would shake the foundations of the entire edifice.

Given the spate of recent revelations about revolting bots, the focus on AI safety and the risks to humanity may well be justified, and should be tackled.

But that should not prevent us from fretting about the delicate, interlinked financial structures that underpin the AI boom, and the risks for us all if they crumble.