Amazon has secured a $17.5 billion delayed-draw term loan led by Citigroup, days after breaking the all-time record for a Canadian dollar bond offering, as the company prepares to spend roughly $200 billion on AI infrastructure this year. The deal's structure signals that Wall Street is giving hyperscalers extraordinarily favorable terms, and the debt arms race among Amazon, Microsoft, and Google is accelerating fast enough to raise real questions about what happens if AI revenue doesn't keep pace.
Amazon didn't just raise money this week. It demonstrated, in two consecutive capital markets moves, that the biggest banks on earth are willing to hand over tens of billions of dollars with almost no strings attached. The $17.5 billion senior unsecured term loan, led by Citigroup with JPMorgan Chase, BofA Securities, HSBC, and Wells Fargo rounding out the syndicate, carries no financial covenants, a delayed-draw feature that lets Amazon pull the funds anytime before September 30, 2026, and a three-year maturity from the draw date, as Bloomberg first reported. Borrowers negotiate for years to get terms that loose. Amazon got them because lenders believe the AI infrastructure buildout is the most creditworthy bet in global markets right now.
That bet is enormous. Amazon has committed to roughly $200 billion in capital expenditure in 2026, up from approximately $128 billion last year, a 56% increase that blew past Wall Street's estimates by about $50 billion and makes it the single largest corporate spender in the world. The money is going toward AWS data centers, custom silicon including its Trainium chips, power grid positioning, and the compute cluster underpinning its deep relationship with Anthropic. AWS's AI revenue had already crossed a $15 billion annualized run rate by the first quarter of this year, which gives management cover for the spending. But the sheer scale means Amazon cannot fund this from operations alone, and the debt markets are being asked to carry an increasingly heavy load.
The term loan was preceded, just two days earlier, by what may be the most striking data point of the week: Amazon sold C$14 billion in senior unsecured notes in the Canadian dollar market, the largest corporate bond ever denominated in Canadian dollars. The five-tranche deal, with maturities ranging from three to thirty years, drew more than C$28 billion in orders, meaning investors submitted twice as much demand as Amazon was willing to accept. It broke a record set barely a month ago by Alphabet, which raised C$8.5 billion in the same market. The hyperscalers are now competing not just in cloud market share but in their capacity to absorb the world's fixed-income liquidity.
The absence of financial covenants in the term loan is worth dwelling on. Covenants are how lenders protect themselves when borrowers run into trouble: they typically require the borrower to maintain certain debt-to-earnings ratios or minimum liquidity levels, and a breach can trigger accelerated repayment or renegotiation. Stripping them out entirely is a concession that lenders make when they believe default risk is negligible and the borrower has the leverage to demand it. Amazon has that leverage, and the banks clearly believe AI infrastructure is not a speculative bet but a near-certainty.
The pricing reflects a similar confidence. Interest on the facility runs at Amazon's choice of a floating base rate or term SOFR, with a margin spread of just 0.625% to 0.875% depending on the company's credit rating. For a $17.5 billion facility, that's extraordinarily tight. The delayed-draw structure adds another layer of optionality for Amazon: it can wait and see how conditions evolve before committing, while the banks sit on committed capital. That's a good deal if you're Amazon.
The systemic question the banks aren't asking loudly
The broader picture is harder to ignore. Morgan Stanley estimates that new AI-related debt issuance could reach $570 billion in 2026, with the heaviest activity still ahead. Bank of America forecasts hyperscaler debt issuance specifically hitting $175 billion this year, more than six times the prior five-year annual average. And across the big four, Amazon, Microsoft, Alphabet, and Meta, capital expenditure is on track to consume roughly 100% of operating cash flows in 2026, compared to a ten-year average of about 40%. That gap is being filled by bonds and loans.
The collective capex commitment from the major hyperscalers now exceeds $630 billion for this year alone. That's a number that presupposes AI monetization will ramp steeply and quickly. AWS's $15 billion AI revenue run rate is real, but it represents a fraction of what would be needed to justify infrastructure at this scale within a conventional payback horizon. The argument from Amazon and its peers is that they're building a generational platform, that the returns will materialize over five to ten years and that sitting out the buildout would be the costlier mistake. Most of Wall Street appears to agree.
But consensus is not the same as certainty, and covenant-free lending means the early warning systems that normally flag trouble have been switched off by design. If AI revenue growth stalls while $570 billion in new debt comes due on three to five year schedules, the adjustment will arrive suddenly rather than gradually. Watch two things over the next year: whether Amazon actually draws the full facility before the September deadline, and whether AWS's AI run rate keeps compounding fast enough to justify the next round of borrowing. The banks have made their bet. Now Amazon has to make the math work.
Also read: Mastercard's Agent Pay for Machines puts blockchain infrastructure at the center of the emerging AI transaction economy • Niteshift raises $7 million to be the cloud layer under every AI coding agent, not just the winning one • Jedify raises $24M to build the context layer that gives enterprise AI agents something to actually work with