Warsh says AI’s hyperscalers are part of why your borrowing costs are rising: ‘The competition for capital is real’

The Federal Reserve raised its benchmark rate by 25bps to 3.75%-4.0% as the 10-year Treasury yield reached 5%, increasing borrowing costs for mortgages, companies and the federal government. Fed Chair Kevin Warsh said heavy AI infrastructure financing is contributing to higher yields: major hyperscalers issued $121 billion of U.S. bonds in 2025 versus a $28 billion annual average in 2020-24, while AI-related global debt could approach $570 billion in 2026. Warsh also cited stronger economic growth and the Iran war, although bond investors continue to point to the federal deficit as a primary driver of elevated long-term yields.

Analysis

The investable consequence is not a uniform "AI is expensive" discount, but widening financing dispersion. MSFT, GOOG, AMZN and META can absorb higher all-in funding costs through scale, liquidity and internally generated cash; ORCL and especially CRWV have materially less room for capex disappointments before leverage, refinancing and equity-duration concerns dominate valuation. A sustained steepening of real yields would therefore shift AI leadership from infrastructure spend beneficiaries toward platforms with monetization evidence and lower incremental capital intensity over the next 1-3 months.

The direct Treasury-supply explanation is likely overstated as a standalone driver: corporate issuance is small relative to sovereign duration supply and macro term-premium repricing. But it matters at the margin because AI borrowers had previously traded with quasi-sovereign credit assumptions; normalization of their spreads raises the hurdle rate for every marginal data-center project. The key 6-18 month risk is that capex remains committed while enterprise AI revenue ramps slower than depreciation and interest expense, converting a productivity narrative into lower FCF and multiple compression—most acutely for ORCL and CRWV.

Financials are a conditional beneficiary, not a blanket long. MS gains from debt issuance, hedging and restructuring activity if credit markets remain open; BAC benefits only if asset yields reprice faster than deposit costs and unrealized securities losses do not constrain capital flexibility. A disorderly move higher in the 10-year would reverse the positive bank read-through through AFS marks, tighter lending standards and rising corporate defaults. The falsification point for the issuer-dispersion thesis is renewed tightening in hyperscaler spreads alongside unchanged or rising AI capex guidance and accelerating AI-linked revenue, which would demonstrate that markets still view the buildout as self-funding.

Key Decisions for Investors

• Initiate a 1-3 month pair: long MS / short ORCL, sized beta-neutral. MS captures issuance and volatility activity while ORCL has greater sensitivity to higher funding costs and capex-to-cash-flow pressure. Exit if Oracle raises FCF guidance or net leverage declines meaningfully; reduce if IG credit spreads widen broadly rather than selectively.
• Maintain a defensive AI basket tilt: overweight MSFT and GOOG versus CRWV, using CRWV puts or a short equity hedge over the next 3-6 months. The asymmetry is weakest for CRWV if data-center financing costs rise or customer concentration becomes visible; risk is a new large contracted-capacity announcement that materially extends revenue visibility.
• Do not add broad bank beta solely on higher long rates. Prefer MS over BAC until deposit beta, AFS sensitivity and commercial-credit trends are confirmed in earnings; a sustained bear steepener with stable credit would be the trigger to add BAC.
• Set monitoring alerts for AI issuer CDS/IG spread widening, quarterly capex-to-operating-cash-flow ratios, and any reduction in hyperscaler capex guidance. A 25-50bp widening in AI-linked issuer spreads without a comparable broad IG move is the actionable confirmation for the long-quality/short-levered-AI trade.