6 hours ago
9
Donald Trump’s most immediate concern in demanding Iran reopen the strait of Hormuz may be rocketing US gasoline prices, but if the conflict drags on, higher energy costs will be felt far beyond the pumps.
Systemically higher power prices and fractured supply chains will squeeze industries and consumers worldwide. For the US, one consequence may be to threaten the fragile economics of the AI boom.
Many oil-importing economies, especially in the global south, are having to contemplate outright shortages of oil and its products. Shops in Egypt face curfews, Indonesia has imposed work from home Fridays and the Philippines has declared a national energy emergency.
As a wealthy oil exporter, the US can largely dodge these concerns. However, as the rising cost of filling up US cars illustrates, it cannot completely avoid the global rise in energy costs – which many analysts now believe will persist for months even if the strait reopens within days.
As a result, many companies will be looking anxiously at their cashflow projections. But for a uniquely energy-hungry industry, whose business model is not yet firmly established and whose investments are financed by huge debts, the challenges may be particularly acute.
OpenAI’s Sam Altman made a less than reassuring comparison in February as he sought to play down fears about AI’s environmental impact in the run-up to what is expected to be a mega launch on to the stock market later this year.
“People talk about how much energy it takes to train an AI model – but it also takes a lot of energy to train a human,” he said. “It takes about 20 years of life – and all the food you consume during that time – before you become smart.”
The Bank of England highlighted the potential link between energy costs and the share prices of AI companies in its regular survey of the risks facing the UK financial system last week.
The Bank’s financial policy committee began by pointing out that investors had already been raising questions about the sector before Trump went to war. “Prior to the conflict, increasing debt-financing needs and concerns about whether expected returns on very significant AI-related investments would materialise led to selling pressure,” it said.
“The conflict could increase these concerns, particularly given the energy-intensive nature of the supply chain for key components and the operation of datacentres.”
It was one aspect of a wider warning that the Iran war could exacerbate pre-existing fragilities in markets, given the likelihood that it will “weigh on growth, increase inflation and tighten financial conditions”.
The chief economist of the World Trade Organization, Robert Staiger, has also made the connection between AI and the impact of the conflict, telling me last month that a prolonged period of high energy prices could “crimp” investment in the sector. “The boom is very energy intensive,” he said.
To underline the real-world consequences of a possible retrenchment, in its latest global trade outlook, the WTO calculated that 70% of investment growth in the US in the first three-quarters of last year was in AI-related goods of one kind or another.
The sheer complexity of the financial engineering underpinning the AI investment mega-boom was laid bare in a forensic note by a US law firm, Quinn Emanuel, published last month, which kicked off by noting that the sector’s revenues last year were about $60bn (£45.3bn) and its capital expenditure $400bn.
For those of us old enough to remember the 2008 global financial crisis, it makes sobering reading – off-balance sheet special purpose vehicles feature heavily, as do asset-backed securities.
Essentially, the “hyperscalers” leading the AI charge, and infrastructure providers such as CoreWeave, are borrowing unimaginably large sums as they dash to build out datacentres (although recent analysis by the AI sceptic Ed Zitron suggests real-world projects lag far behind the promises).
The lenders are often private companies such as asset managers, which makes each company’s total liabilities harder for regulators – or even their investors – to track.
There are separate but interconnected concerns about the activities of this burgeoning private credit sector, which regulators, including the Bank of England, have consistently warned about, highlighting their opacity.
In some cases, tech companies have straightforwardly issued bonds. But there are much more byzantine arrangements at play, familiar from the run-up to the Great Crash.
Datacentre operators have been creating off-balance sheet special purpose vehicles, which “own” the vast datacentres and their future rental income – and borrow against them. In some cases these debts are then pooled together, sliced up and resold to pension funds and investment managers.
As older readers may recall, structures such as these can create false comfort that risks are being spread rather than cumulated, and make it vanishingly difficult to work out exactly who owes what to whom.
Quinn Emanuel’s analysts believe that about $120bn in datacentre debt has been moved off-balance sheets in the past two years. And, as they put it: “The deeply interconnected AI ecosystem means that distress at any single node … can propagate across multiple counterparties and financing layers.”
Higher energy costs for an extended period might conceivably be one trigger for such “distress”, while expectations of volatile interest rates and weaker consumer demand – also likely consequences of the Middle East war – are unlikely to help either.
The fundamental question is familiar: can the AI sector ever generate the revenues to justify sky-high valuations?
But surely even modestly higher energy costs could prompt a rethink – which, given the financial wizardry at work, could cascade out across US markets and beyond.
Could this be yet another way in which Trump’s thoughtless onslaught on Iran has unleashed forces he is powerless to control?
