Private credit ·
Reuters column: private credit's AI lending is valued by model, and GPU collateral assumptions may be fragile
In an opinion column published Oct. 29, Altaf Kassam argues that AI-related private loans are valued by their managers, usually quarterly, with models that tend to move more slowly than markets ('smoothed valuations'). He cites OECD estimates that AI-related deals were 34% of global private credit transactions in 2025, twice the five-year average, and a Bank of England estimate that roughly $700 billion of data-centre spending over 2026-28 will be financed by private credit. He notes that the building is typically only around 30% of an AI data centre's cost (JPMorgan Asset Management), with chips and equipment the other 70%, and that H100 rental rates at big cloud providers have fallen 25% to 50% across regions since early 2025 (Silicon Data). The column links the issue to semi-liquid funds that offer redemptions at stated asset value.
Why it matters for deal teams: If GPUs are collateral, the valuation file needs a current rental or resale benchmark for the chips, not the assumption set at closing; that is the evidence a valuation committee or examiner will ask for.
Source: Reuters Open Interest (ROI) column by Altaf Kassam, syndicated by Devdiscourse (opinion) ↗ · more ↗