AlphaMaven AlphaMaven

AI’s Private Credit Boom Has A Pricing Problem

Reported by Finimize
Oct 29, 2026· 3 min read
Save Post LinkedIn Email
Key Takeaways
  • • Private credit is funding a growing share of AI data-center buildout, and because many of these loans do not trade, stress may surface only when quarterly mark-to-model valuations catch up, according to Finimize.
  • • The OECD estimated AI-related deals were 34% of global private credit transactions in 2025, and the Bank of England said roughly $700 billion of data-center spending over 2026-28 could be financed by private credit.
  • • The Financial Stability Board put the total private credit market at $1.5-2 trillion at the end of 2024 and has warned of a first-mover incentive in semi-liquid funds that could leave remaining investors exposed to later write-downs.
  • • JPMorgan Asset Management said buildings may be only about 30% of an AI data center's cost, while Silicon Data reported the hourly rental rate for Nvidia's H100 chip fell 25%-50% across regions since early 2025, creating collateral risk.
  • • Retirement savers are exposed through target-date funds and the UK's Mansion House Accord, which targets 10% of default workplace pension funds in private markets by 2030, so losses may appear late and in steps.
Summary
Private credit is increasingly financing the AI data-center build-out, a shift that Finimize says creates a pricing problem. Because many of these loans do not trade, managers rely on mark-to-model pricing, estimating value from cash-flow assumptions and comparisons to similar deals. While standard practice, this approach can smooth volatility and understate how quickly conditions are changing, meaning investors may only see stress when quarterly valuations catch up.

The scale of the exposure has made these assumptions consequential. The OECD estimated AI-related deals accounted for 34% of global private credit transactions in 2025. The Bank of England said roughly $700 billion of data-center spending over 2026-28 could be financed by private credit, and the Financial Stability Board placed the whole private credit market at $1.5-2 trillion at the end of 2024. The FSB has also warned of a sharp correction that could lead to sizeable losses.

Collateral presents a further risk. JPMorgan Asset Management has said buildings can be only about 30% of an AI data center's cost, with chips and equipment making up the rest. Pricing power for that equipment can move quickly: Silicon Data reported that the hourly rental rate for Nvidia's H100 chip fell 25%-50% across regions since early 2025. Even if borrowers continue paying, strain can emerge through liquidity, as semi-liquid funds offering periodic redemptions may cap withdrawals when too many investors seek cash at once.

Retail and pension exposure adds urgency. Britain's Mansion House Accord targets 10% of default workplace pension funds in private markets by 2030, and in the US, Vanguard said 59% of participants in plans it administers held their entire account in a single target-date fund in 2024. When underlying loans are valued with quarterly models, pension unit values can look smoother than actual sale prices. The FSB's "first-mover incentive" means early redeemers may be paid at model-implied values, leaving any later write-down to investors who remain.

As a result, losses may arrive late and in steps, tied to reporting dates or redemption restrictions, rather than appearing gradually on everyday statements.

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.

Start your free 7-day trial

Sources

1 outlet

AlphaMaven summarised this story from the reporting below. Each link opens the original article.

Finimize AI’s Private Credit Boom Has A Pricing Problem