Amazon Web Services brought in $45,984 million of Amazon’s $201,944 million in third-quarter 2026 sales, about 23%, but $17,773 million of its $28,076 million operating income, about 63%, by our arithmetic from the earnings release Amazon filed with the SEC on 29 October 2026.
This explainer walks through the four parts of the machine the release describes: what AWS sells, why spending runs ahead of cash, the Trainium chips, and how Amazon’s Anthropic stake shows up in its accounts.
A vertical news Short narrated from this piece’s headline and opening line.
Business Pill · Free Cash Flow
An evergreen Business Engineer explainer on free cash flow. It teaches the general idea only and does not discuss Amazon or this quarter.
The key insight: By our arithmetic, AWS supplied about 23% of Amazon’s third-quarter sales but about 63% of its operating income, while Amazon spent about $1.82 on property and equipment for each dollar of operating cash flow.
AWS rents computing, storage and software to other companies over the internet. Amazon reports it as one segment. The release says “AWS segment sales increased 39% year-over-year to $46.0 billion” and that AWS net sales grew 39.3% to “a $184 billion annualized revenue run rate”.
The segment table puts AWS operating income at $17,773 million, against $11,434 million a year earlier. By our arithmetic that is a margin of about 38.7%. Our results piece covered the quarter in full.
AWS share of Amazon’s net sales and operating income in the quarter ended 30 September 2026, from the segment table in Amazon’s Q3 2026 earnings release. Shares are our arithmetic; ‘Stores & other’ is the North America and International segments combined.
Why Spending Runs Ahead of Cash
To sell more computing, AWS has to buy more servers, chips and data centres first. The cash-flow statement shows purchases of property and equipment of $77,259 million in the quarter, against operating cash flow of $42,422 million. By our arithmetic, Amazon spent about $1.82 on property and equipment for every dollar its operations brought in.
Over twelve months, the release says free cash flow “decreased to an outflow of $42.4 billion”, driven primarily by a $94.8 billion rise in property and equipment purchases, and that “This increase primarily reflects investments in artificial intelligence”. Our cash piece covered how Amazon borrowed to close the gap, and our guidance piece covered the roughly $220 billion of 2026 cash capital expenditures Amazon still expects.
Trainium: Amazon’s Own AI Chips
Trainium is the AI accelerator Amazon designs itself, an alternative to buying every chip from Nvidia. Amazon’s second-quarter release, filed on 30 July 2026, said its chips business “Exceeded a $25 billion annual revenue run rate” and that “the two leading AI labs in the world, Anthropic and OpenAI” were “making multi-year, multi-gigawatt commitments” to Trainium.
The third-quarter release adds one software step: Amazon says it “Made it even easier to adopt its custom AI accelerator, Trainium, with TorchNeuron”, which “lets developers run existing PyTorch code (which most already use) on Trainium without rewriting a thing”. PyTorch is a widely used framework for building AI models.
How the Anthropic Stake Shows Up in Amazon’s Income
Amazon owns a stake in Anthropic, the maker of Claude, and records changes in its value below operating income. The second-quarter release said that quarter’s net income “includes non-operating pre-tax other income of $53.4 billion, primarily from our investments in Anthropic”.
The third-quarter income statement shows other income (expense) of negative $703 million for the quarter, and $68,359 million for the first nine months. The release explains that net income fell from a year earlier mainly because of “pre-tax gains of $9.5 billion in third quarter 2025” from the Anthropic investments. Our earlier piece on Amazon’s Anthropic loop covered the commitments running the other way, from Anthropic to AWS.
Bedrock: Where the Models Meet the Customers
Amazon Bedrock is the AWS service that lets companies use AI models from several makers through one account. The release calls it “AWS’s inference service” and says “Fortune 500 companies spending nearly 7x more year-over-year and total Bedrock token volume growing nearly 17x year-over-year”. The models it lists as added include OpenAI’s GPT-6 family and Anthropic’s Claude models. Our Bedrock piece compared those multiples with Google’s and Microsoft’s.
As we read the segment table, AWS is about a quarter of Amazon’s sales and almost two thirds of its operating income, by our arithmetic.
As we read the cash-flow statement, property and equipment purchases of $77,259 million were larger than the quarter’s $42,422 million of operating cash flow; the release says the twelve-month rise in that spending primarily reflects investments in artificial intelligence.
As we read the two releases together, they describe Trainium chips and the Bedrock model service as AWS products, alongside the AWS segment’s growth to a $184 billion annualized run rate.
Amazon Q3 2026 earnings release, 29 October 2026
“This increase primarily reflects investments in artificial intelligence”
Three Implications
FOR READERS OF AMAZON’S RESULTS Amazon’s release reports AWS as one segment and does not split AI revenue from other services, so the AI share of AWS is not in the filing.
FOR THE CHIPS RACE Amazon’s July release said Anthropic and OpenAI made multi-year, multi-gigawatt commitments to Trainium; the October release adds TorchNeuron to run existing PyTorch code on it.
FOR AMAZON’S INCOME STATEMENT Amazon’s July release tied $53.4 billion of second-quarter non-operating income primarily to its Anthropic investments; the third quarter’s other income was negative $703 million.
The framework’s core line: “In the embedding era, value comes from what the software connects to.”
As we read it, Bedrock is Amazon’s version of that idea: a single AWS service that connects companies to models from several makers. The framework was published in December 2025 and does not discuss Amazon’s third quarter.
Update, 29 Oct: Amazon Leo. Amazon’s third-quarter release says the company “remains on track to begin initial service of its Amazon Leo low Earth orbit satellite service later this year” and has signed a strategic agreement with AT&T, which will be the first major U.S. telecom provider to offer Leo connectivity to enterprise and public-sector customers. Jassy’s letter in the release groups Amazon Leo with Grocery, Pharmacy and Zoox among the “newer businesses” he says are “progressing encouragingly”.
What Is Not Established
The release does not split AWS revenue between AI and other services, does not give Trainium revenue for the third quarter, and does not break down the third quarter’s other income by investment. The $25 billion chips run rate and the Trainium commitments are from the July release, not this one. Amazon’s earnings call had not been read for this explainer.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
By our arithmetic from Amazon’s third-quarter release, AWS supplied about 23% of sales and about 63% of operating income, while Amazon spent about $1.82 on property and equipment for each dollar of operating cash flow; the release ties the rise in that spending primarily to artificial intelligence.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Every figure comes from Amazon’s third-quarter 2026 earnings release, filed with the SEC on 29 October 2026 as Exhibit 99.1 to a Form 8-K, or, where marked, its second-quarter 2026 release of 30 July 2026. Shares and ratios are our arithmetic. Plain-language definitions are ours. Amazon’s earnings call had not been read for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
