账上趴着1431亿，亚马逊为何还要为AI借250亿？

TL;DR

• Amazon issued approximately $25 billion in investment-grade USD bonds, with peak market demand reaching around $62 billion.
• The borrowing isn't due to a cash crunch but rather upfront spending on AI data centers, which is compressing free cash flow.
• Related assets: AMZN, MSFT, GOOGL, META, NVDA, and the data center, power, and AI infrastructure chain.

Amazon completed an approximately $25 billion investment-grade USD bond issuance on July 7th. However, as of the end of the first quarter, the company still held roughly $143.1 billion in cash and marketable securities.

These figures seem anomalous. Why would a company with strong operating cash flow and ample liquid assets on its balance sheet issue long-term debt? The answer lies not in a short-term cash shortage but in the payment cadence of AI infrastructure. Data centers, chips, networking, and power capacity require upfront investment, while revenue generation will materialize over the coming years.

This is also why the market is reassessing Amazon. Historically, investors were more accustomed to viewing AWS as a stable cash cow. Now, AI is pushing the cloud business into a more capital-intensive cycle. Amazon CEO Andy Jassy has called AI a "once-in-a-lifetime opportunity," but analysts are more focused on when these investments will translate back into free cash flow.

AI Factories Spend First, Revenue Comes Later

Let's clarify three financial concepts first. Operating cash flow is the money generated from daily business operations. Capital expenditure is the upfront spending on building data centers, buying servers, and deploying networks. Free cash flow is what remains after deducting capital expenditure.

Amazon's problem is not a drying up of operating cash flow. The company's Q1 earnings report shows that operating cash flow over the past twelve months was approximately $148.5 billion, still at a high level. However, during the same period, free cash flow dropped to about $1.2 billion, a 95% year-over-year decline, primarily due to increased spending on property and equipment.

According to the company's previous disclosures and management's guidance, capital expenditure is expected to be around $200 billion in 2026, up from the approximately $131 billion level projected for 2025. Net spending on property and equipment in Q1 alone exceeded $43 billion, primarily directed towards data centers, custom chips, networking equipment, and AI infrastructure.

Therefore, having $143.1 billion in cash and marketable securities does not equate to "no need for financing." This asset base serves as a strategic buffer for acquisitions, investments, operational needs, and navigating uncertainty. AI data centers are long-term assets, making it more rational to match them with long-term debt.

Bond Market Still Trusts AWS as a Credit Anchor

Another signal from this bond issuance is the enduring strength of buyer demand. Peak market orders reached roughly $62 billion, significantly exceeding the approximately $25 billion issuance size. The initial price guidance for the longest tranche, with a maturity of up to 40 years, was set at around 145 basis points (1.45 percentage points) over comparable US Treasuries.

The implication of investment-grade bonds is quite direct. The credit market is willing to lend to Amazon for an extended period at a relatively modest spread. This doesn't prove that AI investments will be successful, but it does indicate that bond investors currently don't view this financing as a credit deterioration event.

The core support for this remains AWS. Cloud business customer stickiness is high, and the visibility of future contracted revenue is relatively strong. Regulatory filings indicate that as of March 31, 2026, future performance obligations extending beyond one year totaled approximately $364 billion, predominantly related to AWS. While not current revenue, this provides a strong clue about future cash collections.

The bond market is buying a timeline: Amazon first finances the construction of AI capacity; then, in the future, AWS generates revenue and cash flow from this capacity through training, inference, enterprise AI services, and its custom chip ecosystem. As long as this chain remains intact, the increase in debt looks more like a capital structure adjustment.

Jassy Bets on the Capacity Window, Stock Market Watches Collections

Jassy's logic is clear. AI represents a platform-level opportunity, and if a cloud vendor misses the capacity window, the cost could far outweigh a short-term decline in free cash flow. Within this framework, the roughly $200 billion capital expenditure is the ticket to securing demand for the next generation of cloud services.

This narrative has a solid foundation. Both generative AI training and inference require immense computing power. The sooner customers lock in capacity, the sooner cloud vendors need to build out data centers. Amazon's development of custom chips like Trainium and Graviton is also aimed at reducing dependence on external GPUs and tightening the integration between hardware and its cloud services.

However, the cautious scenarios outlined by Bloomberg Intelligence and some Wall Street analysts cannot be ignored. Their concern isn't whether Amazon can borrow money, but whether capital expenditure will continue to exceed market expectations and whether free cash flow will remain under pressure through 2026 and 2027.

This highlights the difference between equity and bond investors. Bond investors primarily need to believe Amazon can pay its interest and principal; a spread around 145 basis points is attractive. Equity investors, however, also care about free cash flow per share, profit margins, and valuation multiples. If AI revenue realization lags behind depreciation and rising interest costs, stock price pressure could emerge sooner.

The Valuation Anchor for Cloud Giants Shifts to Capacity Returns

The broader implication of Amazon's bond issuance is that the valuation anchor for tech giants is changing. Previously, the advantage of cloud vendors primarily came from software economies of scale: more customers lead to lower marginal costs and better cash flow. AI is pulling this model back towards a capital-intensive cycle.

This isn't just Amazon's choice. Other mega-cap cloud and platform companies like Microsoft, Google, and Meta are also ramping up AI infrastructure spending, with industry capital expenditure already reaching hundreds of billions of dollars. The differentiator will be which company can most effectively convert this spending into stable utilization rates and high-margin revenue.

For Amazon, debt itself is not the most dangerous variable. What needs close monitoring is whether AWS's AI-related revenue can keep pace with the slope of capital expenditure. If capital expenditure continues to rise in coming quarters without corresponding improvements in contract conversion, cloud revenue growth, and profit margins, the market will demand a higher risk premium.

Free Cash Flow Will Determine the Market's Verdict on This Debt

This $25 billion bond issuance has brought Amazon's AI investment cycle squarely onto its balance sheet. The company can currently finance at a relatively low cost, and the credit market is willing to underwrite it. However, issuing debt only solves the maturity matching of funds; it does not automatically solve the problem of generating returns on capacity.

The verification points will come in the earnings reports. Investors need to watch whether capital expenditure guidance continues to be revised upwards, whether AWS revenue growth accelerates, whether profit margins hold, and whether free cash flow recovers from its low point. If these metrics fail to improve over time, the market's discussion around Amazon will shift from "Can they build it?" to "If they build it, can they make money?"

The most brutal aspect of the AI infrastructure cycle is that building capacity too early strains cash flow, while insufficient capacity means missing customer demand. Amazon has chosen to secure the funding upfront. Now, over the next few years, it must prove that these AI factories can efficiently turn computing power into free cash flow.