Amazon has signed a $17.5 billion loan agreement with a syndicate of five banks, the latest move in a borrowing spree that has now pulled in roughly $31.5 billion of new financing in 48 hours. Bloomberg first reported the deal, which lists Citigroup, JPMorgan Chase, Wells Fargo, HSBC, and BofA Securities as lenders. The structure is a delayed-draw term loan, meaning Amazon can pull funds when it wants rather than taking the full amount at signing.
The loan landed two days after a separate $14 billion Canadian bond sale, and Amazon has so far described the use of proceeds only as 'general corporate purposes,' according to Reuters. Neither filing earmarks the money for a specific data center, chip order, or acquisition. But the timing tracks with an industry-wide pattern: the biggest spenders on AI infrastructure are now leaning on debt and equity markets to keep capex on schedule.
Amazon is not alone in tapping capital markets at this scale. About a week before Amazon's loan, Alphabet announced an $80 billion stock sale, which the company said would 'fund its investments in a balanced way while retaining a healthy balance sheet.' Meta has lined up a $30 billion bond sale, the largest in its history. Three of the four US hyperscalers have now raised tens of billions inside a single quarter.
Key facts
- 01Amazon signed a $17.5B delayed-draw term loan with Citigroup, JPMorgan Chase, Wells Fargo, HSBC, and BofA Securities.
- 02The loan follows a $14B Canadian bond sale two days earlier, taking Amazon's new financing to $31.5B in 48 hours.
- 03Alphabet plans an $80B stock sale; Meta has lined up a $30B bond sale, its largest ever.
- 04Amazon told lenders the proceeds are for 'general corporate purposes,' per Reuters.
The mechanics of a delayed-draw loan matter here. Amazon does not pay full interest on the $17.5 billion until it draws the cash, which lets the company secure committed bank capacity without inflating near-term interest expense. For a buildout where chip deliveries, site permits, and power interconnects slip on their own timelines, that flexibility is the point. It also signals that Amazon expects to keep deploying capital well beyond what current operating cash flow funds.
The amounts are striking even by Silicon Valley standards. Amazon's $17.5 billion loan alone is larger than the entire 2025 venture funding total for many sectors outside AI. Stacked against Alphabet's $80 billion equity raise and Meta's $30 billion bond, the combined late-cycle financing across just three companies is well over $125 billion — capital that will flow primarily into Nvidia GPUs, custom silicon, data center shells, and the power infrastructure to run them.
The pressure on customer spend is consistent with what the buyer side is doing. AI Chat Daily reported this week on Ramp data showing the top 1% of US firms now spend $7,500 per employee each month on AI tools, a level that, if it holds, helps justify the hyperscaler buildout on the revenue side. Whether that spend rate broadens to the rest of the enterprise market is the open question underwriting every one of these debt deals.
Bond and loan markets have absorbed the supply without obvious strain so far. Investment-grade demand for hyperscaler paper has remained deep because the issuers are among the most creditworthy borrowers in the world, with operating cash flows large enough to service the new interest load comfortably. The risk is not default — it is duration. These are multi-decade infrastructure bets being funded with debt whose payback depends on AI revenue ramps that have not yet fully materialized.
That is the question investors and analysts are increasingly asking. The debate is no longer whether the spending is necessary to stay in the frontier model and inference game. It is whether the returns will justify the capital stack being assembled. Amazon Web Services, Google Cloud, and Azure all report AI-related revenue in the multi-billions and growing fast, but the disclosure granularity is thin and the margin profile of inference at scale is still being established.
Skeptics point to the gap between announced capex and visible AI revenue at the unit level. Hyperscaler capex guidance for 2026 sits well above $300 billion combined, and a meaningful share of that has to clear hurdle rates within five to seven years to keep credit ratings stable. If AI demand growth slows or pricing compresses faster than depreciation schedules assume, the same debt deals being celebrated today as prudent will be re-examined as overextension.
Amazon's move underscores that the AI infrastructure cycle has entered the phase where balance-sheet capacity, not engineering talent, becomes the binding constraint. The hyperscaler with the cheapest cost of capital and the longest committed credit lines wins the next round of GPU allocation, power contracts, and customer commitments. Amazon just bought itself $17.5 billion of optionality on top of $14 billion of fresh bond proceeds — and the message to Alphabet, Meta, and Microsoft is that the spending ceiling is still going up.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
