How AI Debt is Entering Retirement Funds

Key Takeaways

• Debt raised to build AI data centres is being sold on to big institutional investors.
• A single portfolio can end up exposed to the same AI spending twice over.
• Australian data centre operators raised at least A$35 billion in 2026, mostly borrowed.
• Strong tenants and good security still matter; this is not a repeat of 2008.

How AI debt reaches institutional investors

Debt funding the AI buildout is being packaged for pension funds and other institutional investors, according to a report by Americans for Financial Reform Education Fund and the Open Markets Institute. Released on 16 October 2026, the report draws a comparison with subprime mortgages (what caused the GFC): exposure to speculative lending can spread well beyond the original borrowers once their debt is sold through the financial system (Americans for Financial Reform Education Fund and Open Markets Institute, 16 October 2026).

A retirement portfolio could own technology shares while also lending, through another investment, to the data centres those companies use. Both positions may suffer if expansion plans are cut. The investor has bought different assets, but some of the money still depends on the same customers and spending forecasts.

The structure behind the lending

Researchers at the BIS have described how these arrangements work. A separate entity builds or buys a data centre, raises debt from institutional investors and receives payments from a technology company under a long-term lease or capacity agreement. The technology company may also provide guarantees. Much of the borrowing can remain outside its balance sheet, which the Bank for International Settlements describes as "shadow borrowing" (BIS Quarterly Review, Box A, 16 March 2026).

The debt is then serviced from the project's income. If a tenant cannot meet its commitments, lenders may face losses and guarantees may be called. Banks can also be exposed through funding lines. A project that needs refinancing could come under pressure even while its facilities remain busy, if new borrowing costs more than the business can afford.

There are sound reasons to lend against a fully leased facility with a financially strong tenant. The detail of each deal matters, including the lender's claim over the assets and the terms of any guarantee. The subprime comparison is a warning about how risk can be distributed and obscured. It is not evidence that every data centre loan is poor quality or that a repeat of 2008 is inevitable.

What Australia's borrowing figures show

Australia's borrowing figures show how quickly the sector has grown. A September research note published by the Reserve Bank estimated that the operators it examined had raised at least A$35 billion during 2026 to early September, up from A$24 billion across 2025. Debt supplied 85% of the new funding (RBA Insight 2026-07, 23 September 2026).

Most local borrowing came from syndicated bank loans. The note found that securitisation had played a larger role in the United States, where operators raised US$25 billion in 2025 by issuing securities backed by future data centre rents. It also found that Australian data centre financing remained relatively small within domestic capital markets.

AI may deliver substantial economic benefits while some of the companies financing it produce disappointing returns. A lender needs sufficient cash to be paid on time. Even a useful, heavily used facility can be a poor investment if it costs too much to build or its income falls short of the assumptions used to finance it.

What it means for investors

Correlated exposure is hard to see from a portfolio statement. Shares, credit funds and infrastructure holdings can all trace back to the same AI spending assumptions. Physical metal held through Ainslie Bullion, or in tokenised form through Gold Silver Standard, sits outside the AI spending cycle, which is the reason some Australian investors hold it alongside their growth assets.

_______________________________________________

This article is general information only and does not constitute financial advice. Past performance is not indicative of future results. Always conduct your own research or consult a licensed financial adviser before making investment decisions.