Amazon

Amazon reports three segments. North America and International hold the stores, seller fees, ads and Prime. AWS is the cloud. In Q3 AWS brought in 23% of sales and 63% of operating profit, and it grew 39% against 6% for the North America stores.

Ads are the hidden profit line. Amazon doesn't report ad profit, because ads sit inside the retail segments. But ad sales of $19.6B in Q3 were nearly twice the $10.3B the two retail segments earned combined. Ads need no warehouse or truck, so we infer they cover much of the store's thin margin.

The store's profit is recent. North America kept 4.2% of sales as operating profit in 2023 and 6.9% in 2025, and 7.7% in Q3 2026. International went from a $2.7B loss to a $4.8B profit between 2023 and 2025. Lower delivery costs and ads growing faster than sales drove that.

Each piece faces a different rival. AWS leads cloud with 28% of spending to Microsoft's 20% and Google's 15% in Q2, though both are growing faster. Walmart's ads grew 38% in its July quarter to Amazon's 11% in Q3, which Amazon says Prime Day timing held back. Shopify's merchants grew orders 32% in Q2.^{20,21,22,27,7,8,36,37}

In 2023 Amazon split its U.S. warehouse network into eight regions, so most orders ship from nearby instead of across the country. The pandemic had doubled the network in about two years and made it costly to run. By mid-2023, 76% of deliveries were filled within the region, touches per package fell 20% and miles traveled fell 19%.

It showed up in the numbers. Shipping costs rose 7% in 2025 to $102.7B, while store and seller sales rose about 9.5%. Each dollar of sales carried less delivery cost, which is a large part of why North America's margin climbed.

Q2 broke the pattern. Shipping costs rose 19% to $27.9B, faster than the 15% rise in store and seller sales. Amazon blamed higher fuel and line-haul rates. Q3 did not repair it: shipping rose 11% to $28.3B against a 5% rise in store and seller sales, though paid units grew 12% and Prime Day's move held sales down. Amazon keeps widening same-day grocery delivery, now in 2,300 U.S. cities.

Rivals play it differently. Walmart ships online orders from its stores and grew online sales 23% in its July quarter. Shopify owns no warehouses; it sells software to brands, and its merchants' orders rose 32% to $115.6B.^{34,35,22,21,2,7,8,36}

Amazon bought Israeli chip designer Annapurna Labs in 2015 for a reported $350M. That team now makes Graviton processors, Nitro networking cards and Trainium, Amazon's AI chip. The chips business passed a $25B yearly sales pace in Q2, more than doubling from a year earlier. Jassy says it would be about $50B if it sold chips like a standalone chipmaker.

The pitch is price. Amazon says Trainium2 gives about 30% better price-performance than comparable GPUs, and Trainium3, shipping since early 2026, is 30% to 40% better again. Those are Amazon's numbers, not independent tests. Each chip AWS designs is one it doesn't buy from NVIDIA, and cheaper chips help AWS hold margin while depreciation climbs.

Demand is large and narrow. Amazon reported over $225B of Trainium revenue commitments in April. Anthropic has committed to up to 5 gigawatts and OpenAI to about 2 gigawatts, ramping in 2027. Amazon plans to sell Trainium outside AWS from late 2027. The order book and the backlog's concentration are largely the same bet.

Anthropic is also hedging. Its S-1 lists Google's TPUs and NVIDIA and AMD GPUs alongside Trainium. It committed $114.5B to Google through 2033, more per year than its $110B AWS minimum through 2036.^{33,28,1,2,23,11,38}

Backlog is revenue customers have signed for but AWS hasn't delivered yet. It reached $578B in September, up from $496B in June, with an average of 6.1 years left to run. Amazon doesn't break it down by customer. Two contracts are public: OpenAI's $138B in total and Anthropic's minimum of $110B through April 2036. Together that is about 43% of the backlog by our arithmetic, down from about half in June.

Anthropic looks able to pay. Its S-1, filed October 26, shows $16.3B of revenue in the first half of 2026, an operating loss of only $0.7B, and $131.7B of cash. Its AWS minimum works out to about $11B a year. The contract can't be canceled, and if Anthropic uses less, it pays Amazon the difference.

OpenAI is the real exposure. It forecasts $36B of revenue this year and $278B of cash burn through 2030, and on its own plan it runs through its latest $122B raise by 2028. Its AWS contract averages about $17B a year. Amazon has also put $50B into OpenAI, so it is partly funding the bill it expects to collect.

If a lab can't pay, the cost lands on Amazon fast. It buys the servers before the revenue arrives, and AWS depreciation, the yearly write-down of that equipment, rose to $9.2B last quarter from $5.6B a year earlier. Idle servers keep costing that money. The cushion is that Amazon said in July that demand would outrun supply through 2027, so freed capacity would likely find other buyers, maybe not at the same price.^{21,23,29,18,2,37}

Amazon has put about $18B into Anthropic since 2023: $8B of convertible notes, which are loans that turn into shares, and $10B of preferred stock. At September 30 it carried them at $238.7B, $146.2B for the notes and $92.5B for the preferred. That equals about a quarter of Anthropic's last private price of $965B, but it isn't the same as owning a quarter.

The IPO filing sets the terms. The notes turn into up to 1.84 billion nonvoting shares when Anthropic lists. Amazon's stake is capped at 12% of Anthropic's capitalization unless Amazon waives the cap, so any excess stays as notes and converts later. Amazon also expects a lock-up, a period after the IPO when it can't sell.

The stake swings reported profit, not the business. Q2 net income included $53.4B of gains, mostly from Anthropic, but Q3 had almost none, and EPS fell to $1.89. The notes carry a $140.3B unrealized gain that sits outside earnings for now; earlier conversions moved $4.5B of such gains into 2026 earnings. Anthropic's S-1 shows the mirror image: a $72.4B first-half charge.^{21,23,25,26,1,12,37}

Amazon trades at 24 times the earnings analysts expect over the next 12 months. That ratio is the multiple: what investors pay today for each dollar of a year's earnings. That is cheaper than it has been 80% of the time over the past five years, when the middle was 42.

Analysts expect earnings to grow about 29% the year after, against about 16% for the S&P 500. The S&P 500 trades at 19 times. The premium is a bet that Amazon's earnings keep growing faster than the market's for years, not just next year.