McKinsey has, for instance, estimated that $US5.2 trillion will be needed by 2030 for data centres if they are to keep up with the demand for AI computing power. Morgan Stanley has estimated that, between 2025 and 2028, AI-related capital expenditures will be almost $US3 trillion, with roughly half that provided by external capital – new equity and debt.
Increasingly, the balance in the funding for AI will shift from the initial emphasis on internal cash generation and, for the smaller players equity raisings and vendor financing, to debt. That’s already occurring.
McKinsey has estimated that $US5.2 trillion will be needed by 2030 for data centres if they are to keep up with the demand for AI computing power.Credit: Bloomberg
Even the biggest of players, like Google’s parent Alphabet, Meta and Oracle (which has a debt-laden balance sheet already) are tapping debt markets for their AI investments, and we’re starting to see the emergence of off-balance-sheet funding – funding raised within special purpose vehicles secured by specific AI assets rather than the issuing company’s balance sheets – and securitisation of data centre debt.
The combination of frothy asset values and leverage is a combustible one. In this instance, given all the uncertainties around the AI demand and revenue paths and the sheer size of the numbers involved, the risks are magnified.
It’s also a concern that debt is being arranged and packaged against data centres that are reliant on chips that could have a life cycle of as little as three or four years, being made redundant as new generations of chips are developed. There will be high levels of recurrent expenditure required to maintain the centres and their value.
Loading
For the hyperscalers, those risks are manageable. Mark Zuckerberg put it starkly earlier this year when he said, from Meta’s perspective, the risk of over-investment was preferable to being too late to a technological transformation, saying it was possible that Meta could end up misspending “a couple hundred billion dollars” but the risk was higher that it could be “out of position” on the most important technology in history.
For companies without the cashflows and balance sheets of the mega techs to absorb the consequences of a misjudgment of the commercial value of AI relative to their spend or of the timing of the realisation of that value, the question marks over AI are of existential consequence.
As the amount of debt linked to the various strands of AI development mount, those question marks taken on wider significance.
Last month, the Bank of England’s blog, in a discussion of the implications of a fall in AI-related asset valuations for financial stability, referred to the potential for a re-evaluation of future revenue and earnings projections and a subsequent fall in AI-related asset prices.
Morgan Stanley has estimated that, between 2025 and 2028, AI-related capital expenditures will be almost $US3 trillion, with roughly half that provided by external capital – new equity and debt.
It said these could include underwhelming progress in AI capabilities, or user adoption or a below-expectation ability of AI companies to monetise their AI applications, saying the speed of AI progress and its economic impacts were highly uncertain.
It also referred to potential bottlenecks to progress. Most likely was power, it said, but training data and AI chip production could also be factors.
“Financial stability consequences of an AI-related asset price fall could arise through multiple channels. If forecasted debt-financed infrastructure growth materialises, the potential financial stability consequences of such an event are likely to grow,” the bank said.
The AI boom was primarily an equity story until this year, which meant that a fall in AI-related asset prices would not necessarily lead to severe financial stability.
Loading
But past episodes have demonstrated that hidden leverage can exist within the financial system, commodity markets could be affected (each megawatt of data centre power capacity uses between 20 and 40 tonnes of copper) and a fall in AI asset values could adversely impact US growth, where AI investment has been an out-sized driver, it said.
For the moment, discussions about an AI bubble are largely confined to the sharemarket and to the paper losses that would flow if it burst.
As debt assumes a larger role in the funding of the sector, however, and as it is spread among the various channels in the financial system – corporate bond investors, private credit, private equity and banks among them – the threat to system stability from a significant mismatch between the accelerating growth in investment in AI and in the revenue streams it generates will grow.
