Amazon Q3 2026 results: AWS grows 39% to $46 billion and 2026 capex stays at about $220 billion
Amazon reported third-quarter 2026 results on October 29, and the numbers put Amazon Web Services at the center of the company's AI story more clearly than in any quarter so far. AWS segment sales rose 39% year over year to $46.0 billion, up from $33.0 billion in the third quarter of 2025, while total net sales increased 12% to $201.9 billion. Operating income climbed to $28.1 billion from $17.4 billion a year earlier, and AWS alone generated $17.8 billion of it.
The other side of the ledger is the cost of building AI capacity. Amazon spent $77.3 billion on purchases of property and equipment in the quarter, against $35.1 billion in the same period of 2025, and trailing-twelve-month free cash flow turned into an outflow of $42.4 billion. The company said it continues to expect 2026 cash capital expenditures of approximately $220 billion, confirming that the pace of data center, chip and power investment it set earlier in the year has not changed.
Chief executive Andy Jassy framed the quarter around AWS demand. In the earnings release he said that AWS's investment in "functionality and infrastructure is working", cited a $184 billion revenue run rate for the business, and said Amazon projects AWS revenue growth rates of over 40% in both 2027 and 2028. Those projections are company guidance about future years, not reported results, and Amazon did not publish the assumptions behind them.
··········
THE THIRD QUARTER IN NUMBERS.
The headline figures below come from Amazon's earnings release and the segment tables attached to it. Percentages are year over year unless stated otherwise.
........
Metric | Q3 2026 | Q3 2025 | Change |
Net sales | $201.9B | $180.2B | +12% |
Operating income | $28.1B | $17.4B | +61% |
Operating margin | 13.9% | 11.4% | +2.5 points |
Net income | $20.7B | $21.2B | -2% |
Diluted EPS | $1.89 | $1.95 | -3% |
AWS net sales | $46.0B | $33.0B | +39% |
AWS operating income | $17.8B | $11.4B | +55% |
Purchases of property and equipment | $77.3B | $35.1B | +120% |
........
The drop in net income despite much higher operating income has an accounting explanation that Amazon spells out in the release. The third quarter of 2025 included $9.5 billion of pre-tax gains on Amazon's investment in Anthropic, which inflated last year's comparison base. Stripped of that effect, the operating line is the better guide to how the business performed, and it improved substantially.
Amazon also flagged two factors that weighed on the reported growth rate. Foreign exchange had an unfavorable impact of $0.9 billion on net sales, and Prime Day moved into the second quarter in most large countries this year. The company said that, excluding the Prime Day shift, growth would have been about 400 basis points higher. That timing effect matters mostly for the retail segments, not for AWS.
··········
AWS: ACCELERATING GROWTH AND A $578 BILLION BACKLOG.
On the earnings call, as summarized in Amazon's own post about the quarter, Jassy said AWS grew 39.3% year over year, its sixth consecutive quarter of acceleration and its fastest growth in nearly five years. The calculation from the segment table confirms the figure: $45,984 million against $33,006 million is a 39.3% increase.
Profitability moved in the same direction. AWS operating income of $17.8 billion on $46.0 billion of sales implies an operating margin of about 38.7%, compared with about 34.6% a year earlier. AWS produced roughly 63% of Amazon's total operating income in the quarter while accounting for less than a quarter of net sales, which explains why investors read the company's AI strategy primarily through the cloud segment.
The most forward-looking number disclosed is the backlog. Amazon said AWS backlog stands at $578 billion, up by triple digits year over year, and described it as well diversified. Backlog represents contracted future revenue rather than revenue already earned, and Amazon did not break down how much of it is tied to AI workloads or to individual large customers. The company also said core compute and storage revenue grew at its fastest rate in several years, an indication that AI demand is pulling ordinary cloud infrastructure along with accelerators.
Amazon gave a vivid but unquantified description of the build-out: it is installing an AI accelerator about every 14 seconds, and customers are consuming capacity almost as quickly as it is added. Jassy said AI operating margins are improving rapidly, while acknowledging that memory costs and depreciation from capital expenditures weigh on them. He also said inference accounts for well over half of AWS's AI revenue and is still growing, alongside what he called substantial training revenue. Amazon did not disclose a dollar figure for AI revenue specifically.
··········
TRAINIUM, GRAVITON AND THE CUSTOM CHIP BUSINESS.
Amazon's in-house silicon received unusual attention in this quarter's disclosures. The company said its chips business is growing at triple-digit percentages year over year and that revenue commitments for both Graviton CPUs and Trainium accelerators have each risen more than 20x year over year. It did not publish revenue figures or unit volumes for either chip family, so the scale of the business in absolute terms remains undisclosed.
For Trainium, Amazon said short-term capacity on prior-generation instances is selling at a premium compared with the first half of the year, and that it intends to offer Trainium outside AWS starting in late 2027. That would be a notable change in distribution model for a chip that has so far been available only as an AWS cloud service. The company also highlighted TorchNeuron, an open-source project that lets developers run existing PyTorch code on Trainium without rewriting it, which addresses one of the main practical barriers to moving workloads off NVIDIA GPUs.
Graviton's case is less obvious but technically interesting. Amazon argued that AI is a tailwind for its Arm-based CPUs because post-training, reinforcement learning and agent tool use run largely on CPUs rather than accelerators. Agent workloads spend much of their time calling tools, running code and moving data, which are CPU-bound tasks. Amazon claims up to 30-40% better price-performance for Graviton than alternatives, a vendor figure without a disclosed benchmark baseline, and said backlog for prior-generation Graviton instances more than doubled quarter over quarter after the Graviton 5 launch. It also said customers using its custom silicon saved more than $2 billion in the third quarter alone, again without detailing the methodology.
··········
BEDROCK, AGENTS AND THE MODELS NOW ON AWS.
The release lists new AWS agreements with Meta, NVIDIA, OpenAI, Oracle and Qualcomm, among others, and describes rapid growth on Amazon Bedrock, the managed service that gives developers access to third-party and Amazon models. Amazon said Fortune 500 companies are spending nearly 7x more on Bedrock year over year and that total Bedrock token volume grew nearly 17x. Jassy called Bedrock the fastest-growing multi-billion-dollar service in AWS history.
The model catalog expanded considerably during the quarter. According to the release, Bedrock added OpenAI's GPT-6 Astra, GPT-6 Sol, GPT-6 Luna and GPT-6.1 Sol plus two cybersecurity models, Anthropic's Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5, Moonshot AI's Kimi K3, Grok 4.7 and Z.ai's GLM 5.3. Amazon also added built-in web search for OpenAI models on Bedrock and launched a public preview of Bedrock Managed Agents powered by OpenAI. OpenAI's Codex CLI has since added GPT-6.1 Terra to its Amazon Bedrock model catalog, a model covered in our article on GPT-6.1 Terra and its API pricing.
On the agent side, Amazon said Bedrock AgentCore has tens of thousands of customers, who deployed twice as many agents in the third quarter as in the previous three quarters combined. The company released the Strands harness, which it says reduces agent task costs by 32% through token efficiency, previewed Strands Box as an open-source sandbox for defining agent behavior rules, and open-sourced Kiro Crew, a multi-agent workspace. AWS Continuum, an AI security service, became generally available and is being integrated into Claude Code, Codex and Kiro, with Capital One, IBM, Infor, Luminate and Robinhood named as customers. Amazon also announced the acquisition of DuckLabs, the company behind the DuckDB analytical database.
··········
WHAT CHANGES FOR ALEXA+ USERS AND AWS CUSTOMERS.
For consumers, the most concrete changes concern Alexa+. Amazon said it expanded Alexa+ to Australia, India and Japan during the quarter and made it available for free to U.S. customers with compatible Fire TV devices. It launched 34 new devices across Echo, Kindle, Fire TV, Ring, Blink and Bee, including the Alexa+ hardware described in our coverage of the new Echo lineup and its prices. The release does not mention availability of Alexa+ in the European Union, so users there should not read the expansion as covering their region.
Amazon also reported that Alexa for Shopping was used by more than 400 million customers over the last 12 months, with interactions up 570% year over year and more than 100 new capabilities added. On the business software side, Amazon Quick, the company's AI assistant for work, expanded into Outlook, Word, PowerPoint and Excel and can keep running tasks when a laptop is closed. Sellers gained Seller Assistant workflows and a plugin for Amazon Quick and Anthropic's Claude, and advertisers gained an Amazon Ads Agent and Full-Funnel Campaigns for all U.S. advertisers.
For developers and companies building on AWS, the practical signals are about capacity and choice. The broader Bedrock catalog means OpenAI, Anthropic, Moonshot, xAI and Z.ai models can be used under one AWS account, billing and governance setup. The premium pricing of short-term capacity on older Trainium instances indicates that accelerator supply remains tight, which teams planning large training or inference jobs should factor into reservations. Amazon also said Lambda passed a $1 billion annualized revenue run rate and launched Lambda Web Functions.
··········
CAPITAL SPENDING, CASH FLOW AND FOURTH-QUARTER GUIDANCE.
The scale of investment is the main cost of this growth. Trailing-twelve-month purchases of property and equipment reached $215.2 billion, compared with $120.1 billion a year earlier, an increase of about 79%. Operating cash flow over the same period rose 29% to $168.3 billion, but not by enough to cover the spending, which is why free cash flow moved from an inflow of $14.8 billion to an outflow of $42.4 billion. Amazon attributes this primarily to a $94.8 billion year-over-year increase in net purchases of property and equipment, tied to its AI investment.
For the fourth quarter, Amazon guided net sales of $240.0 billion to $247.0 billion, or growth of 12% to 16% compared with the fourth quarter of 2025, assuming an unfavorable foreign exchange impact of about 60 basis points. Operating income is expected between $28.5 billion and $33.5 billion, compared with $25.0 billion a year earlier. The company expects advertising revenue to grow in the high teens year over year in the fourth quarter. In the third quarter, advertising services revenue was $19.6 billion against $17.7 billion a year earlier, an increase of about 11% on the release's figures, and subscription services revenue was $14.0 billion against $12.6 billion.
........
Q4 2026 guidance item | Amazon guidance | Q4 2025 comparison |
Net sales | $240.0B to $247.0B | Growth of 12% to 16% |
Operating income | $28.5B to $33.5B | $25.0B |
Advertising revenue | Growth in the high teens | Year over year |
2026 cash capex | About $220B | Full-year expectation unchanged |
........
The guidance assumes no energy derivative remeasurement impacts and no additional business acquisitions, restructurings or legal settlements. Beyond AI, the release notes that Amazon Leo, the company's satellite internet service, remains on track for initial service later this year with an AT&T agreement in place, and that Zoox now offers paid robotaxi rides in Las Vegas and operates in 12 U.S. cities.
··········
WHAT THE QUARTER ESTABLISHES AND WHAT REMAINS OPEN.
The reported figures establish that AWS is growing faster than at any point in nearly five years, at higher margins than a year ago, and that it now generates most of Amazon's operating profit. They also establish that Amazon is willing to accept negative free cash flow on a trailing basis in order to keep building AI capacity at a pace of roughly $220 billion of cash capital expenditures this year, and that demand for Bedrock, AgentCore and the company's own chips is rising by multiples rather than percentages.
Several important points remain undisclosed or are forward-looking claims. Amazon has not published an AI-specific revenue figure, the share of the $578 billion backlog that comes from AI customers, or absolute revenue for Trainium and Graviton. The projection of AWS growth above 40% in 2027 and 2028, and Jassy's remark that Amazon may have two trillion-dollar revenue businesses within five to ten years, are management expectations rather than results. The claimed savings from custom silicon and the price-performance figures for Graviton are vendor statements without published test conditions. What the quarter shows with certainty is that Amazon's AI bet is now visible in both directions on the income statement and the cash flow statement, and that the next several quarters will test whether backlog converts into revenue fast enough to justify the capital already committed.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org
