AI Companies' Debt Now Equals 68% of New Long-Term U.S. Treasury Borrowing This Year, JPMorgan Finds

JPMorgan estimates six major AI-related issuers raised roughly $320 billion of debt in 2026, including SPV-backed data-center obligations, with about $303 billion in 10-year-equivalent duration—equal to 68% of new long-duration U.S. Treasury borrowing. Goldman Sachs independently estimates about $300 billion of completed AI-related issuance and forecasts roughly $340 billion of senior hyperscaler and chip issuance in 2027 before additional structured financing. The aggregate balance-sheet picture remains manageable for cash-rich hyperscalers, but Oracle is the key credit risk: fiscal 2026 capex was $55.7 billion versus roughly $32 billion of operating cash flow, it is projected to have negative free cash flow through 2029, and its shares are down 17.91% YTD.

Analysis

The key transmission channel is not solvency for the cash-rich platforms; it is the cost of capital. Persistent long-duration corporate supply can widen technology credit spreads and raise the discount rate applied to terminal-value-heavy AI equities, even if policy rates fall. Over the next 1-3 months, this is primarily a relative-value issue: companies funding capex internally retain flexibility on buybacks and acquisitions, while externally financed capacity expansion becomes more sensitive to each incremental turn in spreads.

ORCL is the weakest link because its AI buildout requires capital-market access before the associated cloud revenue and cash conversion are proven. A spread widening would create a reflexive problem: higher financing costs reduce the economics of incremental capacity, which can force lower capex guidance or delay revenue recognition, pressuring both earnings estimates and the equity multiple. The relevant falsifiers are sustained improvement in Oracle free-cash-flow conversion, signed backlog converting into reported OCI revenue, and stable or tighter ORCL long-dated spreads through its next major funding window.

MSFT, GOOGL, META and AMZN should ultimately gain share if smaller or more leveraged AI infrastructure providers retrench, but their equities are not immune: debt-funded capex can crowd out repurchases and make investors demand clearer AI monetization milestones. NVDA faces a second-order risk over 6-18 months if customer financing constraints—not GPU demand—become the binding constraint on data-center orders. Banks benefit near term from underwriting, syndication and structured-finance fees, but the upside is contingent on distribution remaining orderly; widening spreads would shift risk from fee income toward balance-sheet and pipeline exposure.

Consensus appears too focused on aggregate issuance and too little on dispersion. Broad hyperscaler credit stress is unlikely absent a macro shock, but the market can still reprice ORCL sharply and penalize the most capex-intensive AI narratives before headline earnings weaken. Watch 10-year Treasury term premium, long-dated BBB/IG technology spreads, Oracle funding announcements, and any reduction in hyperscaler capex or buyback guidance.

Key Decisions for Investors

• Initiate a 3-6 month pair: short ORCL versus long MSFT, sized beta-neutral. The trade isolates funding and execution risk from broad AI demand; target a further 10-15% relative move, with a stop if ORCL demonstrates positive free-cash-flow inflection or its long-dated credit spread materially tightens versus MSFT.
• Buy 6-9 month ORCL put spreads rather than outright puts ahead of major financing, earnings, or capex-guidance events. This limits premium exposure if the credit concern remains a 2027 issue; avoid chasing after a sharp implied-volatility spike.
• Maintain an underweight in NVDA versus AMZN/GOOGL for the next quarter only if customer financing indicators deteriorate. Escalate the underweight if Oracle or other leveraged cloud providers defer capacity additions; reverse if NVDA reports resilient forward demand with no incremental customer-financing concessions.
• Use a modest long-duration-rate hedge through TLT puts or a short TLT position against concentrated AI equity exposure over 1-3 months. The thesis fails if term premium and long-end yields decline despite continued corporate supply, indicating demand is absorbing issuance without a clearing-rate reset.
• Keep BAC, GS and JPM as watch-list beneficiaries rather than directional longs: add only after disclosed underwriting/markets revenue confirms fee capture without a rise in loan commitments, securities inventory, or credit-loss provisions.