How Amazon makes its money
Amazon reports its figures in three segments. North America and International bundle online retail, the marketplace for third-party sellers, Prime subscriptions and logistics. The third segment is AWS, short for Amazon Web Services. AWS rents out computing power, data storage and AI services to companies. Customers pay for what they use instead of building their own data centers.
A fourth source of income runs across the segments: advertising. Sellers and brands pay to have their products appear near the top of search results. According to the report on the 2025 financial year, advertising services brought in around $68.6 billion. This division is growing faster than retail itself.
In total, Amazon generated revenue of around $716.9 billion in 2025. Net profit was $77.7 billion. Both figures come from the same official earnings release.
Profit is not made where revenue is made
Retail delivers most of the revenue but only a small part of the profit. The North America segment generated revenue of $426.3 billion in 2025. Operating profit came to $29.6 billion, a margin of around seven percent. The international business came to $161.9 billion in revenue and only $4.7 billion in profit.
AWS shows the opposite picture. In 2025, the cloud accounted for $128.7 billion in revenue, just under 18% of the group. Yet it delivered $45.6 billion in operating profit. That is around 57% of total operating income, at a margin of about 35%.
This imbalance is the most important point if you want to understand the stock. Anyone who buys Amazon is, in economic terms, mainly buying a cloud provider. Retail provides reach, customer data and the basis for the advertising business. The stock market valuation, however, depends heavily on how AWS develops.
Why I read the latest figures particularly closely
At the end of July 2026, Amazon reported a quarterly profit of $62.6 billion. I always look at headlines like this twice. The report on the second quarter of 2026 contains a one-off effect of $53.4 billion before tax. This is a valuation gain on the stake in Anthropic. It is not income from ongoing operations, and no money flowed.
Even without this effect, the quarter was strong. Revenue rose by 20% to $200.6 billion. Operating income grew by 43% to $27.5 billion. AWS rose by 37% to $42.2 billion and earned $16.6 billion in operating profit. Book profit and operating profit therefore tell two different stories. I keep them strictly apart before I assess anything.
AI is the fuel for AWS
The AI business of AWS rests on several layers. Companies rent computing capacity to train and run AI models. Through the Bedrock platform, they access ready-made models from various providers. On top of that, Amazon develops its own AI chips under the name Trainium. They are meant to reduce dependence on expensive graphics processors from other manufacturers.
The link to Anthropic is central here. According to its own announcement from April 2026, Amazon has now invested $13 billion. Up to $20 billion more can follow when certain milestones are reached. In return, according to a joint announcement, Anthropic plans to spend more than $100 billion on AWS technology over ten years. The agreement covers access to up to five gigawatts of computing capacity with Trainium chips. The Claude models already run on more than one million Trainium2 chips today. More than 100,000 customers use these models through Bedrock.
The deal serves two purposes for Amazon. It secures AWS a large customer that is committed for the long term. And it gives Amazon’s own chips a demanding buyer that drives their further development.
Three questions before a possible purchase
First, I ask about the business. Is AWS still growing with a high margin, and is the advertising business keeping pace? To answer this, I look at several quarters instead of a single announcement. It also matters whether the company meets its own forecasts. For the third quarter of 2026, Amazon expects revenue of $197 to $202 billion.
Then comes the trend. Even a strong company can be a weak stock if expectations have run ahead of it. So I check whether the price is trading above its long-term moving averages and whether these lines are rising. This does not replace a valuation. It only shows me whether I would be acting against the prevailing direction.
Finally, momentum. After steep rises, a stock is often overstretched in the short term. The Relative Strength Index helps me with this assessment. Only when business, trend and momentum fit together do I consider a decision ready to be tested at all.
The main risks
The investment burden is huge. According to CNBC, Amazon has raised its planned capital spending for 2026 to around $220 billion, from around $200 billion before. One reason is memory chips that have become more expensive. According to the quarterly report, free cash flow over the last twelve months was minus $7.6 billion. If AI demand cools, depreciation will weigh on profits for years to come.
Competitors are investing just as aggressively. According to CNBC, Microsoft and Alphabet are also expanding their cloud capacity with sums in the hundreds of billions of dollars. Today, no one can say who will earn the best return on this spending.
Regulation remains an open case. In an antitrust lawsuit, the US regulator FTC accuses Amazon of unfairly protecting its market power in retail. According to the specialist news service MLex, the trial is scheduled for March 2027. Back in September 2025, according to NPR, Amazon paid $2.5 billion in a settlement over Prime subscriptions.
The retail margin remains thin. A margin of a few percent leaves little buffer for labor costs, tariffs or weak consumer spending. Small shifts in retail can therefore move the group result noticeably.
Book gains can reverse. The Anthropic effect from the second quarter is based on a valuation, not on payments. If the valuation of the AI company falls later, this will weigh on reported profit in the same way.
How I put record figures into context
With record announcements, I first strip out the one-off effects. Then I compare the segment margins over several quarters. Finally, I read management’s forecast and its statements on capacity. According to Fortune, Amazon CEO Andy Jassy said that demand exceeds even the expanded capacity until at least 2027. That sounds good. But it describes expectations, not secured revenue.
I do not use leveraged products on the stock for this kind of assessment. Leverage can force a position to close before the thesis can be confirmed or disproved. For a company in the middle of an investment cycle, that is an avoidable additional risk.
My conclusion on Amazon stock
The shorthand “Amazon equals online retailer” is misleading. On the stock market, Amazon is above all a position in AWS. The cloud delivers most of the profit and, thanks to AI demand, is once again growing significantly faster. The stake in Anthropic and its own Trainium chips tie the group closely to this market.
On the other side are enormous investments, thin retail margins and an ongoing antitrust case. Whether the billions spent on data centers pay off will be decided over years, not in a single quarter.
No recommendation follows from this. For Amazon, I answer the same three questions as for any stock. How is the business developing? Which way is the trend pointing? How overstretched is the current move? This sequence protects against headlines that only tell half the truth.