Business à l'instant3Ajouter aux favoris

Ken Griffin's team is the first major hedge fund to quantify the systemic exposure from AI infrastructure debt - and the number is large enough to matter for credit markets.
In plain terms: Citadel projects that debt issued specifically to finance AI chip infrastructure will reach $500 billion by 2028. That's a hedge fund - not an advocacy group - putting a concrete figure on a risk that regulators and analysts have been describing only qualitatively.
The story: This matters because Citadel's estimate implicitly prices the systemic risk of the AI capex cycle. The debt issuance is being driven by hyperscalers, colocation providers, and now sovereign compute programs, all financing GPU clusters against future revenue that remains unproven at scale. The earlier S&P downgrade of Oracle (capex $90-95B in FY27) and the BRI's warnings about structural parallels to 2008 now have a headline number attached.
The $500B figure is for chips specifically - it excludes the real estate, power infrastructure, and cooling capex that accompanies each cluster build.
Under the hood: Griffin's concern is contagion through the credit channel. If one major borrower (a hyperscaler or cloud provider) faces a credit event, the interconnected vendor-investor structure (Nvidia supplying, then taking equity stakes in customers) creates the kind of circular exposure that amplified prior cycles.
So what: Watch for spread widening on AI-linked debt issuances in H2 2026. The Citadel estimate functions as a market ceiling that credit analysts can price against. If actual issuances approach $200-300B this year, expect rating agencies to revisit their AI capex theses.
Article produit par intelligence artificielle, relu sous contrôle éditorial humain.
Connectez-vous pour rejoindre la discussion.
Fifty billion a year? That’s a heavy bet on AI staying profitable. What if demand collapses before the debt matures?
500 billion is a scary number, but isn’t this just capitalism riding another hype wave? If the tech pays off, who really loses?
If the demand for AI chips drops, we’re looking at a domino effect-supply chains, jobs, even tech hubs relying on that cash flow. The true cost might be higher than the interest alone.
La dette de l'IA : capex, notations et risque de contrepartie