Amazon has gone back to the debt markets for at least $25bn, its largest bond sale of the year and the clearest sign yet of how much the company is prepared to borrow to keep pace in the artificial intelligence race. The offering is split across eight tranches with maturities running from 2029 out to 2066.
The money is earmarked for the thing swallowing cash across the whole industry: data centres, custom silicon, and the physical scaffolding that AI now demands. It follows a pattern set by Amazon's record Canadian dollar bond earlier in the year, part of a borrowing run that has now topped $70bn since the start of 2025 across dollar, euro and Swiss franc deals.
Much of that spending flows through Amazon Web Services, where the company is racing to add capacity for customers training and running large models. A growing share is also going into its own Trainium chips, which Amazon has pitched as a cheaper alternative to buying Nvidia hardware at scale.
Investor demand was there, though it cooled as the terms firmed up. Orders peaked at around $62bn before the banks managing the sale trimmed the spread on offer, leaving the final book at roughly $41bn, or about 1.6 times the size of the deal.
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol' founder Boris, and some questionable AI art. It's free, every week, in your inbox. Sign up now!
Amazon told the underwriters it does not plan to issue any more debt this year, a detail that reads as reassurance to a market watching the sector stack on leverage. It is the kind of guidance that matters when investors are trying to gauge how much more paper is coming.
The borrowing has an obvious cause on the balance sheet. Amazon has guided to roughly $200bn in capital spending for 2026, the most of any hyperscaler, and that bill has already compressed its free cash flow to a fraction of what it was a year earlier.
The company is far from alone in reaching for debt to plug the gap. The four largest US tech firms have collectively guided to more than $650bn in AI capex this year, a figure that increasingly outstrips what even their vast operating cash flows can cover.
Not everyone reads the appetite as bottomless. Analysts noted that demand for this sale looked muted next to some earlier offerings, a hint that bond buyers are turning choosier as AI-linked debt floods the market.
That caution is showing up elsewhere in the financing chain. Deals such as Cipher's junk bonds, raised to build an Amazon data centre in Texas, are testing how far investors will stretch to fund the boom at the riskier end.
Amazon structured the offering as a mix of fixed and floating-rate notes, giving it flexibility across a curve that reaches four decades out. The longest tranche, due in 2066, is the sort of ultra-long paper usually reserved for the most creditworthy names.
Amazon can still borrow at those terms because it remains one of the highest-rated corporate issuers in the market, which keeps its funding costs low even as the totals climb. That gap between what Amazon pays and what smaller rivals pay is now part of the story of who can afford to keep building.
The sale also lands in a market already thick with AI-linked paper from Meta, Oracle and a string of data-centre developers. Each new deal tests the same question, which is how much more debt investors will absorb before they start demanding a higher price for it.
For now the sale hands Amazon a fresh cushion as it heads into a stretch of heavy spending with no let-up in sight. Whether the market stays this willing on the next trip is the question now hanging over the entire sector.