Key Takeaways
- • Altaf Kassam, founder of Reference Portfolio Partners and former head of Investment Strategy at State Street Investment Management, argues in a Reuters opinion column that AI infrastructure financing through private credit is obscuring risk because holdings are valued infrequently by their managers.
- • The OECD estimates AI-related deals made up 34% of global private credit transactions in 2025, twice the five-year average, according to the column.
- • The Bank of England estimates roughly $700 billion of data-centre spending over 2026-28 will be financed by private credit, against a private credit market the FSB estimated at $1.5 trillion to $2 trillion at the end of 2024.
- • The column cites Blue Owl Capital, one of the largest data-centre lenders, twice capping redemptions from its private credit funds at 5% this year after heavy withdrawal requests.
- • Retail savers, including UK workplace pension members and US 401(k) participants, are increasingly the marginal buyers of these less transparent assets.
Summary
A Reuters opinion column by Altaf Kassam warns that much of the AI infrastructure build-out is being financed through instruments that do not trade, including private credit funds, infrastructure vehicles and joint ventures. While hyperscalers such as Meta Platforms, Alphabet and Amazon have tapped public debt markets, a significant share of sector funding occurs privately. Because these holdings have no market price, managers typically value them quarterly using expected cash-flow models and comparable transactions. Kassam argues that this process can produce "smoothed valuations" that understate volatility and make private credit appear to offer higher returns with lower risk.
The scale of AI lending is cited as a central concern. The OECD reckons AI-related deals made up 34% of global private credit transactions in 2025, twice the five-year average. The Bank of England estimates that roughly $700 billion of data-centre spending over 2026-28 will be privately financed, according to the column. Kassam notes that the Financial Stability Board estimated the overall private credit market at $1.5 trillion to $2 trillion at the end of 2024.
Kassam identifies collateral as a fragile model assumption. Citing JPMorgan Asset Management, he writes that the building typically accounts for only about 30% of a data centre's cost, with chips and equipment making up the remaining 70%, and lenders increasingly accept those chips as security. He cites data from Silicon Data showing the hourly rate for Nvidia's H100 has fallen 25% to 50% across regions since early 2025, with smaller specialist providers charging less still. He acknowledges counterarguments: the largest AI tenants are highly creditworthy and private credit default rates remain low, and patient, hold-to-maturity investors may not need frequent markdowns.
Kassam argues the liquidity mismatch is already visible. Blue Owl Capital, one of the largest data-centre lenders, twice capped redemptions from its private credit funds at 5% this year after heavy withdrawal requests. He cites the FSB's concept of a "first-mover incentive" in semi-liquid funds that redeem at stated value, where early redeemers are paid at par and remaining investors bear any shortfall once marks adjust.
The column also highlights growing retail exposure. In Britain, the largest workplace pension providers have pledged under the Mansion House Accord to allocate 10% of default funds to private markets by 2030, and US 401(k) plans are being opened to alternatives. The FSB estimated retail investors held around 13% of global private credit in 2024. Kassam cites Financial Conduct Authority data showing more than 90% of UK workplace pension savers are in default strategies, and Vanguard data showing 59% of participants in its workplace plans held a single target-date fund in 2024. He concludes that the marginal buyer of AI's least transparent debt could increasingly be savers who do not read fund reports.
The scale of AI lending is cited as a central concern. The OECD reckons AI-related deals made up 34% of global private credit transactions in 2025, twice the five-year average. The Bank of England estimates that roughly $700 billion of data-centre spending over 2026-28 will be privately financed, according to the column. Kassam notes that the Financial Stability Board estimated the overall private credit market at $1.5 trillion to $2 trillion at the end of 2024.
Kassam identifies collateral as a fragile model assumption. Citing JPMorgan Asset Management, he writes that the building typically accounts for only about 30% of a data centre's cost, with chips and equipment making up the remaining 70%, and lenders increasingly accept those chips as security. He cites data from Silicon Data showing the hourly rate for Nvidia's H100 has fallen 25% to 50% across regions since early 2025, with smaller specialist providers charging less still. He acknowledges counterarguments: the largest AI tenants are highly creditworthy and private credit default rates remain low, and patient, hold-to-maturity investors may not need frequent markdowns.
Kassam argues the liquidity mismatch is already visible. Blue Owl Capital, one of the largest data-centre lenders, twice capped redemptions from its private credit funds at 5% this year after heavy withdrawal requests. He cites the FSB's concept of a "first-mover incentive" in semi-liquid funds that redeem at stated value, where early redeemers are paid at par and remaining investors bear any shortfall once marks adjust.
The column also highlights growing retail exposure. In Britain, the largest workplace pension providers have pledged under the Mansion House Accord to allocate 10% of default funds to private markets by 2030, and US 401(k) plans are being opened to alternatives. The FSB estimated retail investors held around 13% of global private credit in 2024. Kassam cites Financial Conduct Authority data showing more than 90% of UK workplace pension savers are in default strategies, and Vanguard data showing 59% of participants in its workplace plans held a single target-date fund in 2024. He concludes that the marginal buyer of AI's least transparent debt could increasingly be savers who do not read fund reports.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.