Amazon Q3 Earnings: AWS Growth, AI Spending ⚡
Get the latest Amazon Q3 earnings: AWS growth, Amazon AI spending, and AI infrastructure investment—plus what the $42.4B free cash flow deficit means.
Amazon's AI Bet Is Growing Faster Than Its Cash Flow
Amazon's latest quarterly results lay out two very different stories. AWS is growing at its fastest pace in almost five years, and customers are snapping up new AI capacity almost as quickly as Amazon can install it. Meanwhile, free cash flow swung from positive $14.8 billion to negative $42.4 billion over the 12 months to September.
That mismatch is the core question behind Amazon's AI push: can the company turn outsized demand for compute into returns that make sense for this level of spending? This won't be answered by revenue growth alone. What matters is whether the new infrastructure becomes profitable fast enough, whether AWS can keep its growth momentum, and how much financial flexibility Amazon has while it builds.
This article walks through the Q3 numbers, what Amazon is building, how the spending changes the company's financial risk, and what all of it says about the rest of the business. The goal isn't to treat high capex as automatically good or automatically bad. It's to separate what the demand looks like today from what still has to be proven.
What do Amazon's Q3 earnings say about AI demand?
The clearest signal in Amazon's Q3 report is the pace of AWS growth. Revenue at Amazon Web Services rose 39% to $46 billion - its fastest growth rate in almost five years. AWS operating income climbed to $17.8 billion from $11.4 billion a year earlier. In other words, the AI demand isn't showing up only in experimental products. It's landing inside a mature cloud business.
AWS now has an annual revenue run rate of $184 billion and a reported backlog of $578 billion. A backlog isn't the same thing as already-earned revenue - it reflects contracted work expected to be delivered over time. And its value depends on customers following through on the capacity they've committed to. Still, it gives Amazon a real base of demand as it expands.
On the earnings call, Andy Jassy put the operational pressure into plain terms: "We're installing an AI accelerator about every 14 seconds, and almost as fast as we install capacity, customers are gobbling it up." That's both the opportunity and the constraint. Amazon isn't saying customers aren't interested. It's saying the company has to build physical capacity quickly enough to keep up with that interest.
Amazon also expects AWS revenue to grow more than 40% in both 2027 and 2028. That's not a promise - it's a forecast, and it sets a high bar. At that pace, Amazon has to do more than sell access to AI models. Customers need accelerators plus data-center capacity, networking, storage, and software that lets teams actually use the infrastructure in production.
Read growth and profitability together
AWS operating income matters because it suggests the cloud unit isn't simply buying revenue at any cost. The $17.8 billion operating income shows AWS can make real profit while it expands. But a quarterly margin doesn't tell you the full return on the latest infrastructure build-out. The economics of those investments will become clearer as equipment is deployed, used, and depreciated over time.
Amazon's company-wide results looked strong too: revenue rose 12% to $201.9 billion and operating income increased to $28.1 billion from $17.4 billion. Operating margin came in at 13.9%, the highest the company has reported. That helps explain why management feels comfortable investing. It doesn't remove the need to judge whether the next wave of spending will earn a return worth the risk.
The market reaction was about expectations, too
Amazon's shares fell as much as 7% in extended trading after the results. Revenue and earnings per share came in below analyst forecasts. That reaction doesn't contradict AWS's momentum. It reflects the gap between a solid operating quarter and what investors had already expected - along with lingering questions about the cost of sustaining future growth.
If you're reading this as an investor, the takeaway isn't "AI demand is strong, so the strategy is proven." It's more specific: customers are consuming AWS capacity quickly, and AWS is growing fast. The next test is whether Amazon can keep adding capacity without letting spending run ahead of profitable demand.
Why is Amazon spending so much on AI infrastructure?
AI services require a different build-out than adding conventional cloud servers. Training and running AI workloads depend on specialized processors and the surrounding infrastructure to keep them running - power, networking, and data systems included. Amazon is leaning into that physical base because customers want access now, and because capacity takes time to plan, buy, and install.
In Q3, Amazon spent $77.3 billion on property and equipment, up from $35.1 billion a year earlier. Purchases of property and equipment rose by $94.8 billion over the 12 months to September, and the company says most of that increase was driven by AI. For the year, Amazon expects cash capital spending of about $220 billion (the same figure it set in July), and Jassy said spending in 2027 would be "meaningfully more."
This is a trade-off with real consequences. Data-center assets are long-lived and expensive. Build too slowly and you risk leaving customer demand unserved and giving competitors an opening. Build too quickly and you can end up with costly equipment that's underused - or returns that fall short of the company's cost of capital. Amazon appears willing to accept more near-term financial pressure to avoid being short on capacity while demand is rising.
Chips are part of the product strategy
Amazon is also trying to shape the economics of the infrastructure through its own chips. It plans to sell its Trainium AI chips outside AWS in late 2027. Earlier this month, it released TorchNeuron for Trainium, which lets developers run PyTorch code on the chips.
This gives Amazon more than one lever. Its own silicon could help it offer customers more choices in price and performance, and potentially reduce how dependent it is on buying every accelerator from outside suppliers. But custom chips only help if developers can adopt them without a ton of engineering work. Software compatibility, performance, availability, and support determine whether customers actually make the switch.
TorchNeuron matters here because developer tools can reduce friction between a chip and an existing workflow. The real question isn't whether the tool exists - it's whether teams can move workloads (or build new ones) without taking on disproportionate migration and maintenance costs. That kind of adoption takes time across a customer base.
Capacity is an operating commitment, not just a purchase
When management says it is installing an accelerator every 14 seconds, the number points to deployment pace, not directly to productive capacity. Even after an accelerator is installed, it has to be connected to the rest of the system, made available to customers, and kept busy enough to justify its cost.
For companies buying cloud services, more capacity can mean more options and easier access to AI compute. It can also complicate procurement decisions. Teams should validate workloads against real operating requirements: latency, availability, data handling, software compatibility, and total cost. Faster infrastructure doesn't automatically make every AI workload commercially useful.
Jassy has talked about the opportunity in huge terms, saying Amazon could have two businesses with $1 trillion in revenue within five to ten years. That's ambition, not a plan you should use as a baseline. The more useful evidence is the near-term stuff: AWS revenue, operating income, backlog, capacity utilization, and what it actually costs to build the next generation of infrastructure.
What does the cash-flow decline mean for Amazon?
Amazon's free cash flow was negative $42.4 billion over the 12 months to September, compared with positive $14.8 billion a year earlier. This is the biggest counterweight to the growth story. Amazon says AI investment drove most of the decline, while operating income increased significantly. In other words, the pressure isn't just a sign of weaker operations. It's a reflection of cash going into equipment and other property before that spending shows up as full economic returns.
That distinction matters, but it doesn't erase the cash-flow reality. Capital spending uses cash upfront, before the assets have time to generate what they're ultimately meant to produce. If new capacity becomes productive and supports profitable customer growth, the cash drain can be a rational cost of building a bigger business. If demand slows, pricing drops, or utilization disappoints, the same spending becomes harder to justify.
Amazon raised $48.1 billion of long-term debt during the quarter. A filing after the results also showed the company can borrow up to $80 billion against the AI hardware it buys. Taken together, those details suggest the build-out is being funded through operating cash, existing financial resources, and borrowing capacity. That doesn't, by itself, point to financial distress. But it does make the funding plan more important to watch as spending ramps up.
What investors should watch beyond the headline deficit
A single free-cash-flow number isn't enough to judge a long investment cycle. But it's also not helpful to treat capex as inherently good. A better approach is to track a small set of indicators across multiple quarters:
- AWS revenue growth: Does demand stay strong as the revenue base gets larger?
- AWS operating income: Is AWS growing profitably, not just selling more?
- Capital spending: Is investment continuing to rise, and at what speed?
- Free cash flow: Does the cash deficit level off or keep widening as equipment is deployed?
- Backlog and customer uptake: Are commitments turning into delivered, paid-for services?
Read them together. Rising capex alongside accelerating AWS revenue and operating income looks different than rising capex with flat demand. And an improving free cash flow can be a positive - but only if it isn't coming from underinvesting in capacity needed to serve customers.
Management confidence is not a return metric
Jassy said Amazon can fund its plan and sees a clear line of sight to attractive returns. That's helpful context because it frames the investment as deliberate rather than reactive to one quarter's results. Still, "clear line of sight" is exactly the kind of phrase you should pressure-test against actual performance as the assets come online.
The key financial question is whether incremental returns from new AI infrastructure justify both the cost and the risk. Amazon has advantages: a big cloud customer base, an established global business, and existing operating profit. But there's a real trade-off too. Moving early can protect Amazon's ability to meet demand, while moving too far ahead of durable demand can tie up cash and increase financing needs.
How should businesses and investors read the results?
For investors, the Q3 report is a good example of separating operating momentum from investment risk. AWS is growing quickly, and backlog suggests contracted demand extends beyond the current quarter. Company-wide operating income and margin improved as well. On the other hand, free cash flow turned sharply negative, capital spending is high, and management expects higher spending in 2027.
This report doesn't settle whether the AI build-out will deliver attractive long-term returns. It shows demand exists today - and that Amazon is willing to spend to meet it. The remaining uncertainty is economic: what customers will actually pay for these services, how durable their usage will be, and whether AWS can hold onto healthy margins as capacity expands.
Retail adds some useful color. Online store sales increased 2% to $68.5 billion, while advertising revenue grew 11% to $19.6 billion. Amazon said moving Prime Day into the second quarter in most large countries reduced total growth by about four percentage points. That timing effect means you shouldn't take the reported retail growth rate at face value without considering the calendar shift - but the numbers still suggest retail and advertising weren't growing at the same pace as AWS.
Jassy also said Prime Video has a revenue run rate in the tens of billions and is profitable. That comment gives a wider view of Amazon's portfolio, but it doesn't change the central question here. AWS is the growth engine driving these results, and everything else has a different contribution to growth and cash flow.
A practical framework for evaluating the AI bet
If you're assessing Amazon - or making a similar infrastructure decision inside your own business - break it down into three questions:
- Is there demonstrated demand? Look for customer usage, contracted commitments, and revenue, not just market interest or executive forecasts. AWS's growth and backlog are meaningful signals, but backlog still needs to convert into delivered, paid-for services.
- Can the supplier serve demand profitably? Track operating income, utilization, and the cost of adding capacity. Revenue can rise quickly while the economics of the new capacity remain unproven.
- Can the investment survive a slower period? Consider free cash flow, debt, and how easy it is to adjust spending. A plan that only works if demand keeps accelerating has less room for error.
For business buyers, the same discipline applies. Don't build an AI program based on the assumption that more compute automatically creates value. Start with a workload, measure performance and operating costs, and expand only when the service improves a business outcome enough to justify ongoing spend. Amazon's infrastructure may open more doors - but it can't make your use case automatically work.
The holiday quarter will add another data point
For the holiday quarter, Amazon expects revenue of $240 billion to $247 billion and operating income of $28.5 billion to $33.5 billion. Guidance is a range, not an outcome, but it gives investors a near-term way to compare expected retail-season performance with the ongoing infrastructure investment.
The most useful follow-up isn't just whether Amazon beats the forecast. It's whether AWS growth, profitability, and cash spending keep moving in a way that supports the longer-term plan. A strong quarter can confirm demand. It can't prove the full investment cycle will earn attractive returns.
Conclusion: demand is visible, returns are still being built
Amazon's Q3 earnings show a strong mix of signals: AWS demand is accelerating, and Amazon is spending at a scale that's meaningfully reducing free cash flow. Those facts don't cancel each other out. The growth supports the decision to add capacity; the cash-flow decline makes execution and returns harder to ignore.
The right stance is to treat the AI build-out as a high-conviction investment with milestones you can measure, not as a guaranteed win or a reckless spending spree. Watch whether AWS converts its backlog into sustained revenue and operating profit, while capital spending and cash flow stay within a range that makes sense. For business buyers, evaluate AI infrastructure through concrete workloads and unit economics. Demand is a good starting point. Durable returns are what matters next.