The same Wall Street that financed the dot-com bubble is now betting $5 trillion on AI infrastructure, except this time they're doing it with debt instead of equity.
The Summary
- Wall Street is funding a $5 trillion AI buildout through unprecedented debt issuance for data centers, GPUs, and infrastructure while pulling back from software investments
- Oaktree warns massive AI debt could create future cracks in credit markets, particularly in private credit and leveraged loans
- Hidden refinancing risks are emerging as investors continue buying despite rising interest rates
- Credit managers are still finding 8-10% yields, but the question is how long that lasts when the AI debt comes due
The Signal
JPMorgan reports Wall Street has abandoned caution and shifted capital from software to physical AI infrastructure at a scale that dwarfs previous tech buildouts. We're not talking about cloud subscriptions or SaaS multiples. This is steel, silicon, and power grids. Data centers that need financing before they generate a dollar of revenue. GPUs bought on credit before anyone knows what they'll actually produce.
The mechanics matter here. When you finance software, you're betting on gross margins above 80% and capital-light scaling. When you finance data centers, you're betting on utilization rates, power costs, and lease terms. Different risk profile. Different debt service coverage. And right now, lenders are writing checks anyway.
"Massive AI debt issuance could create future cracks in credit markets, especially in private credit and leveraged loans."
Oaktree's credit team sees the warning signs that always appear late in a credit cycle. Spreads tightening while rates stay elevated. Covenant-lite structures becoming standard. Private credit funds competing for deals by loosening terms. The same managers who are finding 8-10% yields today are the ones who will face writedowns tomorrow if refinancing windows close or utilization disappoints.
The refinancing risk is the sleeper issue. Most AI infrastructure debt was issued when everyone assumed rates would fall by now. They haven't. Higher rates may persist longer than markets expect, which means companies that borrowed to build data centers in 2024-2025 will face a refinancing wall in 2027-2028 at rates nobody underwrote for.
Key vulnerabilities emerging:
- Leverage ratios that assume 90%+ data center utilization
- Private credit facilities with floating rates and thin covenants
- GPU financing structured like equipment leases, but for assets that depreciate faster than cars
The Implication
If you're building in AI infrastructure or adjacent to it, stress test your assumptions for a world where your customers can't refinance cheaply. The companies most at risk are the ones in the middle: not the hyperscalers with fortress balance sheets, and not the pure software plays with no capex. The vulnerable layer is data center operators, GPU-as-a-service providers, and infrastructure middleware that took on debt to scale fast.
For investors, the 8-10% yields Oaktree mentions are real, but they're compensation for refinancing risk that isn't fully priced yet. Watch credit spreads in private tech debt and leveraged loan funds with AI infrastructure exposure. When spreads start widening, it won't be gradual.