Amazon filed preliminary terms for a corporate bond offering sized between $37 billion and $42 billion, the third jumbo tech issuance since early March. Meta priced $15 billion in late February. Google parent Alphabet followed with $10 billion mid-March. All three disclosed AI infrastructure as primary use of proceeds.

The Amazon raise will be the largest single corporate bond offering of 2025 if it prices at the upper end. The company's last significant debt raise was $18.5 billion in May 2021, when ten-year investment-grade paper traded near 2.1%. Today's benchmark ten-year IG spread sits at 3.8%, a 170-basis-point premium that Amazon will absorb to fund data-center capacity now rather than wait for rate normalization. The cost of delay—lost training cycles, competitor model deployment, enterprise customer migration—exceeds the cost of capital.

This is not financial-engineering arbitrage. This is capacity procurement treated as strategic imperative. The AI infrastructure buildout has converted Big Tech balance sheets into quasi-sovereign balance sheets. Bond desks are pricing these issuances against duration risk and sector rotation, not default risk. Amazon, Meta, and Alphabet collectively hold $289 billion in cash and marketable securities as of their most recent filings. They are borrowing not because they lack liquidity but because they are pre-funding multi-year capital programs that cannot be funded piecemeal from operating cash without creating quarterly earnings volatility that the market would punish more severely than the debt service.

The capital-markets implication is immediate. Investment-grade corporate bond supply for Q2 2025 is already running 34% above the five-year average for the period. Pension funds and insurance allocators have absorbed the Meta and Alphabet paper without spread widening because the duration and credit quality fit liability-matching mandates. If Amazon prices successfully at scale, it confirms that the bond market will underwrite AI capex at sovereign-equivalent rates for the hyperscalers. That creates a funding-cost moat that smaller model builders and inference providers cannot cross. The gap between those who can borrow at 3.8% for ten years and those who pay 8-11% for venture debt is the gap between building frontier models and licensing them.

Allocators should watch two follow-on events. First, whether Microsoft moves to match this financing within 90 days—the company has been notably absent from the jumbo issuance cycle and holds $104 billion in cash but faces identical capex pressure. Second, whether secondary spreads on the Meta and Alphabet paper tighten or hold after Amazon prints. If they tighten, it signals the bond market is treating AI infrastructure as a new asset class with scarcity value, not just corporate credit.