Warnings about powerful, uncontrolled AI dominated the headlines last week. But even amid the harbingers of the apocalypse, it’s still worth asking whether the entire AI industrial complex is sailing toward a financial iceberg. It will hardly be our most pressing concern if robots are about to take over the world, but a collapse of the AI bubble would have repercussions far beyond the US. Whether technologies must be allowed to progress rapidly and how they can be used. There is ample evidence that AI urgently needs to be regulated, from the ability of Meta’s perverted 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 discussed, including independent analysis of AI models, look like major improvements to the ungoverned status quo. But we must also be alert to the risk that a small number of closely linked mega-companies that have accumulated multimillion-dollar debts expect the State to open a regulatory moat for them. Perhaps a government-backed AI “pause” could prevent cheaper Chinese options from encroaching on Silicon Valley’s market dominance, for example. The first concern is the huge scale of debt issuance being used to finance the frenetic pace of data center deployment by the hyperscalers that build them (Google, Amazon, Microsoft, Meta and Oracle): $132bn (£99bn) this year alone by one estimate. In a world of fragile bond markets with yields on the 10-year US Treasury bond, a global benchmark for borrowing costs, hovering around 5%, the size of these debt piles could be a potential trigger. for a rethinking of the market. The scale of debt issuance being used to finance the frenetic pace of data center deployment by hyperscalers has been estimated at $132 billion. Photograph: Aleksei Gorodenkov/Alamy This is especially true given the second cause for concern: the fact that the “unit economics” of AI remains questionable and is not 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.” He noted that OpenAI has repeatedly reduced its fees to retain customers, for example. An index from research firm Silicon Data that aims to track how much customers pay for a million tokens (the units of data processed by large language models) shows it has more than halved since June, to less than $1. income growth. Anthropic reportedly recently told investors that its “adjusted operating income” was positive; the only problem is that this measure effectively excludes many of its costs. As digital rights advocate Cory Doctorow puts it: “These companies claim they are so great that their profitability can only be measured using a new, secret mathematical way.” Next Day’s Newsletter PromoSome aspects of this edifice may seem familiar to veterans of the global financial crisis: masters of the universe, with a business model that’s so clever that mere mortals can’t be expected to understand it, all underpinned by massive amounts of leverage. A recent, downright scary, research note from financial analyst Groundbreaker made this analogy clear, uncovering a third reason to worry: Simply adding debt doesn’t represent the financial promises underpinning the boom. The $1.5 trillion “start-up wall” that AI labs faced over the next two years, drawing a comparison to when cut-price “teaser” mortgage rates began to dry up in 2007 and 2008. When those rates ended, pushing low-income homeowners into much higher rates, borrowers began defaulting en masse, lighting the fuse of what which became the global financial crisis. According to Groundbreaker’s analysis, in many cases data centers are being built and equipped on take-or-pay contracts, with not a dollar owed until a deadline (often two to three years) is met, and they come to life. Meanwhile, the hyperscaler building the data center books the contract value as expected future revenue, and shareholders applaud. Meanwhile, the buyer (a cutting-edge lab like 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 that the steep jump in costs, as contracts mature and data centers come online, could account for a staggering $700 billion next year and more than $800 billion in 2027. All of that could be fine if revenues continue to soar. Or maybe not so much, if AI end users are not willing to pay enough to cover the costs, perhaps because cheaper options emerge. Technically, this is not debt, but the impact, if the obligations cannot be met in full, would shake the foundation of the entire edifice. Given the spate of recent revelations about nasty robots, the focus on AI safety and risks to humanity may well be justified, and needs to be addressed. But that shouldn’t stop us from worrying about the delicate, interconnected financial structures underpinning the rise of AI and the risks to us all if they crumble.
