Why Is Wall Street Pouring $5 Trillion Into AI Debt While Abandoning Software?

According to JPMorgan's David De Boltz, speaking on Bloomberg Open Interest on May 19, 2026, the AI debt boom is fundamentally different from the software lending boom of the 2010s. "The difference is collateral," De Boltz said. "A data center with power contracts and a GPU cluster is a hard asset. A SaaS company with recurring revenue is a promise." This shift has led Wall Street to allocate record capital to AI infrastructure — an estimated $5 trillion globally by 2030 — while pulling back from software loans, which now carry higher perceived risk due to rising rates and slowing growth. The result is a bifurcated credit market: AI infrastructure debt is oversubscribed; software debt is being restructured.

Why Are Investors Still Buying AI Debt Despite Rising Rates?

De Boltz explained that investors are treating AI infrastructure debt as a "real asset play" rather than a pure tech bet. "When rates rise, most tech debt gets crushed because future cash flows are discounted more heavily," he said. "But AI data centers have power purchase agreements (PPAs) and GPU lease payments that are often indexed to inflation or have built-in escalation clauses." This means the debt service costs are partially hedged. Furthermore, the secondary market for GPUs — particularly Nvidia's H100 and B200 — remains liquid, providing a floor for recovery rates. Bloomberg reported that AI infrastructure debt deals in Q1 2026 had an average coupon of 8.5%, compared to 6.2% for investment-grade corporate debt, yet they were 3.5x oversubscribed. The thesis: even if the AI bubble deflates, the hardware retains value.

Who Wins and Who Loses in the AI Debt Boom?

The winners are clear: GPU financiers like CoreWeave and Lambda, which have raised billions in debt backed by Nvidia hardware. According to JPMorgan's analysis, these firms can borrow at 8-9% and earn 15-20% returns from leasing GPUs, capturing a healthy spread. The losers are traditional software lenders — banks like Silicon Valley Bank (now First Citizens) and specialty finance firms that lent to SaaS companies at low rates in 2020-2022. Those loans are now trading at distressed levels as refinancing becomes impossible. De Boltz noted that "the software lending market is effectively frozen for all but the top 10% of companies." The biggest loser may be late-cycle investors who buy AI infrastructure debt in 2026 without understanding the 2027 refinancing cliff.

What Are the Hidden Refinancing Risks in AI Tech Credit?

De Boltz highlighted a specific risk that most market commentary ignores: the "refinancing mismatch" between the life of AI infrastructure debt (typically 5-7 years) and the useful life of GPUs (3-4 years). "You're financing a 5-year asset with a 7-year debt structure, but the GPU is obsolete in 3 years," he said. This means that when the debt matures, the underlying collateral may be worth significantly less than the loan balance. Bloomberg reported that some AI infrastructure deals include "GPU refresh clauses" that require borrowers to maintain a minimum compute capacity, effectively forcing them to take on more debt to upgrade hardware. This creates a debt spiral: companies must borrow more to keep their collateral valuable. The risk is highest for firms that borrowed in 2024-2025 to buy H100 GPUs, which are already being discounted as B200 shipments ramp up.

What Does This Mean for the Broader Tech Credit Market?

The AI debt boom is creating a two-tier tech credit market that will persist through at least 2028. According to JPMorgan's analysis, AI infrastructure debt will account for 40% of all tech debt issuance by 2027, up from 15% in 2024. This concentration creates systemic risk: if AI demand falters, the entire tech credit market will freeze, not just one segment. The software lending market, meanwhile, will remain depressed, forcing more SaaS companies to pursue venture debt or equity financing at unfavorable terms. The long-term consequence is a consolidation of tech lending among a few large players — JPMorgan, Apollo, and Blackstone — who can afford to hold illiquid AI debt on their balance sheets. Smaller banks and specialty lenders will be squeezed out.

1. By Q3 2027, CoreWeave will restructure at least $2 billion of its debt as GPU prices fall and refinancing becomes unavailable at current spreads.
2. Apollo Global Management will raise a $10 billion distressed AI debt fund by Q1 2028, targeting 20%+ returns from buying impaired AI infrastructure loans.
3. The EU will propose AI infrastructure debt disclosure rules by Q4 2027, requiring banks to report collateral valuations and refinancing risk metrics.

• Wall Street is treating AI infrastructure debt as a real asset class, not a tech bet — but the collateral (GPUs) depreciates faster than the debt matures.
• The 2027 refinancing wall is the single biggest unhedged risk in tech credit, and most CLO investors are unaware of the GPU obsolescence mismatch.
• Software lending is effectively dead for all but the top SaaS companies, creating a funding gap that will accelerate M&A by cash-rich tech giants.
• Distressed debt specialists like Apollo are the true winners of this cycle, not the GPU financiers who will face a reckoning in 18 months.
• The AI debt boom is rational but fragile: it depends on continued GPU demand growth, which is not guaranteed given the cyclical nature of hardware upgrades.