Opinion: Amazon's record profits are increasingly built on borrowed money

Amazon's third-quarter results, released Thursday evening, were by several measures the best in the company's history. Operating income hit $28.1 billion on a 13.9 percent margin, which Amazon's own earnings release filed with the Securities and Exchange Commission calls its highest operating margin ever. Amazon Web Services grew 39.3 percent, its fastest pace in nearly five years, to $46 billion in quarterly revenue. Yet the stock barely moved, and in the hours around the report it was, if anything, soft. The reason is not that Wall Street failed to notice the good news. It is that the good news arrived alongside a balance sheet that looks nothing like the one Amazon had a year ago, because the company's artificial-intelligence buildout has quietly shifted from being funded mostly by its own cash to being funded increasingly by borrowed money.

That shift, not the margin record, is the more important story in this earnings season, and it deserves more scrutiny than it has received.

A headline quarter with a quieter footnote

The topline numbers were genuinely strong. Net sales rose 12 percent to $201.9 billion, a touch below Wall Street's consensus, while earnings per share of $1.89 missed the $1.97 analysts had expected, according to earnings trackers compiled by MarketBeat's rundown of the report. Some of the softness was technical — Prime Day's calendar shift into the second quarter cost the comparison roughly four percentage points of growth, as one earnings recap of the report noted — but it still left a rare miss for a company this size. AWS now runs at an annualized $184 billion and carries a $578 billion order backlog, evidence that cloud and AI demand is not a mirage. CEO Andy Jassy, in comments carried by Amazon's own newsroom, framed the investment as paying off in real time.

"We're installing an AI accelerator about every 14 seconds," Jassy said, describing the pace of Amazon's data-center buildout.

What got less attention is what is paying for those accelerators. Free cash flow, which Amazon itself reports on a trailing-twelve-month basis, swung from a $14.8 billion inflow a year ago to a $42.4 billion outflow, by the company's own account a result of a $94.8 billion year-over-year jump in net capital spending. Capital expenditure for the full year is now guided at roughly $220 billion, and management told investors to expect "meaningfully more" in 2027. A business that generates that little free cash while promising to spend even faster next year has to get the difference from somewhere else.

From operating cash to borrowed cash

It is getting it from debt markets, at a scale that has changed Amazon's financial profile in under a year. According to the company's own quarterly report filed with the SEC, long-term debt stood at $65.6 billion at the end of 2025. By September 30, it had nearly tripled, to $173.1 billion. Over the first nine months of the year, financing activities brought in $111.8 billion in net cash, with $115.1 billion of that coming from new debt proceeds. The sourcing was not subtle:

• A $37 billion multi-tranche bond sale in March, one of several debt offerings tied explicitly to AI infrastructure
• A further $25 billion bond issuance in July
• A $17.5 billion term loan drawn in September

Amazon has also arranged a new secured financing facility of up to $80 billion specifically for technology infrastructure assets, available from early 2027 through late 2028, according to the same filing. None of this is hidden; it is disclosed in detail in regulatory filings that any analyst can read. But the practical effect is that Amazon, long celebrated for generating enormous cash from operations and funding its own expansion, is now running something closer to a capital-intensive utility's financing model: borrow heavily against future demand, build the asset, and hope usage catches up before the bill comes due.

The bet nobody can unwind quickly

The complicating factor is that some of that future demand is itself contractual, and circular in a way that deserves attention. Amazon's filing discloses an AWS compute commitment with OpenAI worth $100 billion over eight years and one with Anthropic worth more than $100 billion over ten years. Those are real contracts with real revenue attached, and they help explain why Jassy's team is confident enough to guide AWS growth above 40 percent for both 2027 and 2028. But they also mean a meaningful share of the revenue justifying Amazon's capital spending comes from a small number of AI labs whose own business models remain unproven and who, by some accounts reported elsewhere in this earnings season, are themselves financing their compute purchases with debt and vendor credit. If AWS's biggest AI customers were to stumble, Amazon would be left holding depreciating data-center assets financed with real, fixed-rate liabilities, not with hypothetical revenue.

This is a materially different risk than the one investors priced through most of the 2010s, when Amazon's capital spending was funded largely from operating cash and any miscalculation mainly cost the company growth, not solvency. A $173 billion debt load carries interest obligations regardless of whether AWS's AI tenants pay their bills on schedule.

The case for not worrying

The strongest objection to this argument is that Amazon's credit is excellent, its interest coverage remains comfortable at a 13.9 percent operating margin, and heavy capital spending ahead of demand is exactly how Amazon built AWS into a $184 billion business in the first place. Bulls can reasonably note that AWS operating margin held at 38.7 percent even while absorbing elevated depreciation, that the $578 billion backlog provides real visibility, and that bond investors clearly agree the risk is manageable, given that Amazon's debt sales have drawn demand well above the amount offered. Jassy's own language about accelerating "capex payback" suggests management believes the return math already works, not that it is a hopeful bet on future returns.

That case is not frivolous, and it may well prove correct. But it rests on an assumption that AI infrastructure demand will keep compounding at close to its current rate for several more years without a pause, even as the capital intensity of meeting that demand keeps rising faster than revenue. Markets that extend that much credit on that assumption are not pricing a sure thing; they are pricing a story, and the muted reaction to a record quarter suggests even generous investors are starting to notice the gap between the earnings headline and the cash-flow reality underneath it.

Investors and regulators alike would do better to track Amazon's free cash flow trajectory and debt-service costs over the next two reporting cycles than to fixate on AWS growth rates in isolation. If free cash flow begins recovering toward positive territory as 2027 capacity comes online and starts billing, the current borrowing will look like the sensible, if aggressive, infrastructure financing that Amazon says it is. If the capital intensity keeps outrunning revenue while debt keeps compounding, the conversation about Amazon's AI spending will stop being about optimism versus caution and start being about leverage versus a slowdown it did not plan for. Thursday's results did not answer that question. They mostly confirmed that the stakes for answering it correctly have gotten much larger, much faster, than most of the headlines about record margins let on.