Capital

Amazon Borrowed $25 Billion for AI Capacity

Amazon raised at least $25 billion in an eight-tranche bond sale to help fund a roughly $200 billion capex year for AI. The order book was pared and Amazon paid up to fill it. For anyone whose stack runs on AWS, the build-out is now financed with debt, and that is both a commitment and a warning.

CapitalBy Servola Tech Desk2026-07-113 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

Amazon Borrowed $25 Billion for AI Capacity

Key takeaways

Amazon raised at least 25 billion dollars in an eight-tranche bond sale on July 7, 2026 to help fund a roughly 200 billion dollar capital-expenditure year, up from about 131 billion in 2025, meaning the AI build-out at the largest cloud provider is now financed with debt, not just cash flow.
Order books peaked near 62 billion dollars but were pared to about 41 billion after underwriters trimmed the yield, and Amazon dangled extra spread to fill the book, a sign that even the strongest issuer is now paying up to borrow for AI.
Across the big cloud vendors the sector expects to spend more than 700 billion dollars on AI infrastructure this year, increasingly on borrowed money; if your stack sits on AWS, treat this as both a commitment signal and a reason to watch pricing and contract durability.

What Amazon put on the balance sheet

On July 7, Amazon sold at least 25 billion dollars of bonds in eight tranches, with maturities running from three to forty years and a mix of fixed and floating rates, tagged for general corporate purposes that include capital spending and debt paydown. Barclays, Goldman Sachs, JPMorgan and Morgan Stanley led the deal. It follows about 54 billion dollars of bonds issued earlier in the year and a 10 billion dollar Canadian offering in June, taking Amazon's 2026 borrowing to roughly 89 billion dollars, and the company said it does not plan to issue more debt this year.

The money underwrites a spending year of a different order. Amazon has guided investors toward about 200 billion dollars in capital expenditure for 2026, up from roughly 131 billion in 2025, most of it data centers and chips for AWS. AWS revenue grew 28 percent in the most recent quarter, and Amazon's own Trainium and Graviton silicon is now projected to generate more than 10 billion dollars a year. The scale is not the surprise. The funding mix is: the largest cloud provider is now borrowing, at size, to keep building.

The tell in the order book

The interesting number is not the 25 billion, it is what happened underneath it. Orders for the deal peaked near 62 billion dollars, then were pared to about 41 billion once the banks trimmed the spread, leaving demand at roughly 1.6 times the size. To get there, Amazon had to offer extra yield relative to where its bonds normally price. A double-A borrower dangling a concession to fill a book is a small signal with a large meaning: the cheap-money phase of the AI build-out is closing, and even the strongest names now pay a premium to finance it.

Amazon is not alone. Alphabet, Microsoft, Meta and Oracle are all tapping debt and equity markets, with the group collectively pointing at more than 700 billion dollars of AI infrastructure spending this year. When that much capital is being raised against a demand curve nobody can yet prove, the market starts asking whether the returns will arrive on the timeline the debt assumes. That question does not have to end badly to matter. It just has to make lenders, and customers, price the risk.

What it means if your stack runs on AWS

For an operator running on AWS, the read is two-sided. The commitment side is real: chief executive Andy Jassy has said demand for AI capacity is outrunning what Amazon can bring online, so the capacity you may be queuing for is being built at speed. But demand outrunning supply is also an allocation-and-pricing statement. Capacity that is scarce and expensive to finance does not get cheaper for the customer first. A build-out funded by 40-year bonds carries a fixed obligation that has to be repaid whether or not the AI demand curve cooperates, and repayment pressure tends to travel downstream into list prices and contract terms.

The practical move is to treat cloud cost as a multi-year durability question rather than a monthly line. Model what your AWS bill looks like if committed-use discounts tighten, read renewal terms for the right to reprice, and keep a credible second option warm even if you never use it. None of this means leaving AWS. It means recognizing that the provider has just put its AI build-out on credit, and that the interest on that credit is one of the inputs that will eventually shape what you pay.

Frequently asked questions

Why is Amazon borrowing when it has strong cash flow?

Because the scale of AI capital spending has outrun what even Amazon's cash flow covers comfortably. The company has guided toward about 200 billion dollars in capex for 2026, most of it data centers and chips, and borrowing lets it fund that build-out while preserving cash, locking in long-dated rates rather than drawing down reserves.

Is the pared order book a problem?

Not a crisis, but a signal. Demand fell from about 62 billion dollars to 41 billion after the banks trimmed the yield, and Amazon offered extra spread to complete the sale. That a double-A issuer had to concede to fill its book suggests the cheap-financing window for AI infrastructure is narrowing, which raises the cost of the whole sector's build-out.

What does this mean for AWS customers?

Read it two ways. It confirms Amazon is building capacity aggressively, which is good if you are waiting on it. But a debt-funded build-out carries fixed repayment obligations, and with the chief executive saying demand is outrunning supply, scarce capacity plus financing cost tends to show up in pricing. Model cloud cost over several years and check your renewal terms for repricing rights.

Amazon borrowing 25 billion dollars to keep building is not a distress signal, it is a statement that the AI build-out has outgrown even the richest cash flows and now runs partly on credit. The tell is that the strongest issuer in the market had to pay up to place the debt, and that cost does not stay with the lender. For anyone whose business sits on top of a hyperscaler, the useful response is not alarm but arithmetic: model the cloud bill over years, read the repricing clauses, and treat the interest on this build-out as one of your own future inputs.

AmazonAWSAI infrastructurebond salecapexdata centershyperscalers

Sovereign Cloud Now Comes With a Score

AWS launched its European Sovereign Cloud in January 2026 with a 7.8 billion euro build. In June the Commission explained its scoring rubric for cloud sovereignty. The label era is over; the grade era has begun.

3 min read

Capital

SK Hynix Raised $26.5B in a Record Nasdaq Debut

SK Hynix priced its Nasdaq ADR debut at USD 149 on July 10, 2026, raising about USD 26.5 billion, the largest US listing ever by a foreign company, and the stock jumped 17 percent on day one. What it means for the memory market and compute buyers.

4 min read

Digital Sovereignty

The EU Council Revived the Rule That Scans Private Messages

The EU Council revived the lapsed rule that lets providers scan private messages with AI. For owners, confidential communication is now a governance choice.

2 min read

← All articles