Google, Meta, Microsoft, and Amazon have collectively spent $112 billion on capital expenditures in the past three months alone to fund artificial intelligence infrastructure. Companies are increasingly turning to complex debt-financing options, including asset-backed securities and off-balance-sheet vehicles, reminiscent of instruments used before the 2008 financial crisis.
The sheer scale of spending has spooked investors, with Meta’s stock tumbling 11 per cent after the company revealed its aggressive capital expenditure plans, whilst tech stocks have sold off this week on overvaluation fears, reports the New York Times.
According to McKinsey, $7 trillion in data centre investment will be required by 2030 to keep up with projected demand. Morgan Stanley estimates that $800 billion in private credit will be needed over the next two years to fund data centres, with special-purpose vehicles becoming a more popular way to structure debt.
About $13.3 billion in asset-backed securities backed by data centres has been issued across 27 transactions this year, a 55 per cent increase over 2024. Investment giant Blackstone is on the cusp of closing a $3.46 billion commercial-mortgage-backed securities offering to refinance debt held by QTS Data Centers, the biggest player in the AI infrastructure market, representing the largest deal of its type this year.
The bonds would be backed by 10 data centres in six markets, including Atlanta, Dallas and Norfolk, Virginia, that together consume enough energy to power Burlington, Vermont, for half a decade. The offering is unprecedented for the commercial-mortgage-backed securities market, where issuance for data-centre-backed deals was just $3 billion for all of 2024.
Special-purpose vehicles
While Meta, Microsoft, Amazon, and Google previously relied on their own cash flow to invest in data centres, they have more recently turned to loans. When Meta structured its $30 billion debt offering for its new data centre in Louisiana, the largest private capital transaction on record, Morgan Stanley arranged the debt to sit in a custom, off-balance-sheet special purpose vehicle, making Meta look healthier on paper.
The Bank of England warned in October that as companies continue to shift from using their own cash flow to amassing debt for data centres, risk will continue to mount. AI stocks now account for roughly 44 per cent of the S&P 500 market capitalisation, up from 26 per cent in late 2022, with these companies frequently trading at valuation multiples that imply high expected future earnings growth.
The cyclically adjusted price-to-earnings ratio of the S&P 500 is close to its dot-com peak, with AI stocks having a median forward 12-month price-to-earnings ratio of 31 times, compared to 19 times for the broader S&P 500 index. These valuation multiples have pushed some US stock indices close to levels seen at the peak of the dot-com bubble.
The Bank of England noted that multiple factors make the transformational economic impact of AI uncertain. At the same time, the physical infrastructure that underpins AI model training and inference is expected to require trillions of dollars of investment over the next five years, a significant share of which is expected to be financed by debt.
“This is a fast-evolving topic, and the future is highly uncertain,” the Bank of England wrote, adding that if hyperscalers are unable to generate enough profit to offset costs related to capital expenditures, systemic risk could enter credit markets.
According to Menlo Ventures, only three per cent of consumers pay for AI-related services, amounting to about $12 billion per year. The share of single-asset-single-borrower securities, where assets inside the bond being sold are all from the same company or a single data centre, is rising, with 13 per cent of all such deals coming from data centres.
Goldman Sachs estimates that 60 per cent of data centre power demand growth through 2030 will need to be met with new capacity, while the International Energy Agency estimates that electricity demand from data centres worldwide will more than double by 2030 to a consumption level higher than that of Japan.
