Amazon conference call: While laying off employees, it makes a massive bet of hundreds of billions on AI. The CEO says, "The faster we increase capacity, the faster we can monetize."
Amazon announced that its capital expenditures will reach $125 billion in 2025 and continue to grow, confidently stating: "The faster we increase capacity, the faster we can monetize." In terms of AI chips, the company is not only collaborating closely with NVIDIA to ensure supply but is also vigorously developing its self-developed chip Trainium, which has grown to a scale of several billion dollars. Regarding the $1.8 billion severance costs due to layoffs, it emphasized that this move aims to reshape the company culture
On one hand, there are the largest layoffs in history, and on the other hand, there is a massive gamble of AI capital worth hundreds of billions of dollars. Amazon still firmly believes that its substantial AI investments will yield rich financial reports.
In the face of market skepticism regarding AI and concerns about slowing cloud demand, Amazon delivered a strong financial report on October 30, with AWS cloud business revenue growth returning to 20% year-on-year, reaching a nearly three-year high, and annualized revenue climbing to $132 billion.
During the earnings call that day, CEO Andy Jassy announced that the company expects its total capital expenditure for 2025 to reach approximately $125 billion, with further increases in 2026, most of which will flow into data centers, power, and chips required for AI. He directly addressed investors' concerns about the potential for overcapacity due to massive investments, emphasizing, "We are monetizing our capacity as quickly as we are increasing it."
Regarding the highly anticipated chip strategy, Jassy clarified Amazon's "two-legged" approach. On one hand, they are developing their own chips, and on the other hand, they are deepening cooperation with NVIDIA, stating, "We have purchased a large number of NVIDIA chips."
Meanwhile, Amazon has recently attracted attention due to an $1.8 billion severance cost. Jassy explained that this move is not purely financially driven but aims to reshape the company culture, reduce management layers, and enable the company to act and innovate quickly like "the world's largest startup" in response to current technological changes.
Wall Street Journal article previously reported that Amazon plans to cut up to 30,000 employees, equivalent to about 10% of its total white-collar workforce, affecting multiple business departments including human resources, cloud computing, and advertising.
Key points from the call:
- Strong AWS Growth: Amazon Web Services (AWS) revenue grew 20.2% year-on-year, reaching $33 billion, marking an 11-quarter high, with annualized revenue reaching $132 billion.
- Increased AI Capital Expenditure: Amazon expects total capital expenditure for 2025 to be around $125 billion, with continued increases in 2026, primarily to support AI and core service demands. CEO Andy Jassy stated that the company is "very aggressively" investing in capacity, and new capacity is being rapidly monetized.
- Self-Developed Chip Strategy Effective: The self-developed AI chip Trainium business has reached a scale of billions of dollars, with Trainium growing 150% quarter-on-quarter, used by major clients like Anthropic for training large models. Amazon also emphasized its deep cooperation with NVIDIA, stating, "We are purchasing a large number of NVIDIA chips."
- AI Services as a New Growth Point: CEO Andy Jassy expects that the AI inference service Bedrock could potentially become a business on par with EC2 in the long term. The newly launched AI agent building service AgentCore has already been downloaded a million times, with companies like Sony and Ericsson adopting it
- Strategic Intent of Layoffs: Regarding the recent layoffs, Amazon stated that the recent $1.8 billion severance cost is not primarily driven by financial or AI factors, but rather to reshape the culture of the "world's largest startup," aiming to reduce hierarchy and enhance efficiency and speed.
- Robust Advertising Business: Advertising revenue reached $17.7 billion, a year-on-year increase of 22%, marking the third consecutive quarter of accelerated growth, with collaborations with platforms like Netflix and Spotify expanding the influence of its DSP platform.
AWS Returns to High Growth, AI Investment "Monetizes as Much as Possible"
Company CFO Brian Olsavsky pointed out that the strong growth this quarter is mainly driven by robust demand for AI and core services, as well as the increased capacity from new launches. By the end of the quarter, AWS's backlog had grown to $200 billion.
Faced with such a massive business scale, Amazon is expanding its infrastructure at an unprecedented pace. Jassy revealed that in the past 12 months, the company has added over 3.8 gigawatts of power capacity, more than any other cloud service provider, and plans to add at least 1 gigawatt more in the fourth quarter of this year. He set an ambitious goal: by 2027, AWS's total power capacity will quadruple compared to 2022.
Regarding the market's most concerned issue of AI investment returns, Jassy gave a very confident answer. He stated that although the company is actively investing, the new capacity is being rapidly monetized. Moreover, the massive investments are based on market demand, with Jassy stating, "We see the demand."
One Hand with NVIDIA, One Hand with Self-Developed Chips
On the much-watched chip strategy, Jassy clarified Amazon's "two-legged" approach.
He emphasized that the company's relationship with NVIDIA is "very deep," and that they have "purchased a large number of NVIDIA chips," while also collaborating with AMD and Intel. This directly addresses market concerns about whether Amazon would reduce cooperation with external partners due to self-developed chips.
At the same time, self-developed chips are playing an increasingly core role. Jassy disclosed for the first time on the call that its self-developed AI training chip Trainium has developed into a "multi-billion dollar annual revenue business," with revenue in the third quarter alone surging 150% quarter-on-quarter.
Currently, the Trainium 2 chip is fully booked, primarily serving a few hyperscale customers. Its core advantage lies in cost-performance, providing 30% to 40% better cost-performance than other GPU options. AI company Anthropic is training its next-generation Claude model on a massive cluster containing nearly 500,000 Trainium 2 chips and plans to expand to 1 million chips by the end of the year.
In addition to hardware, Amazon is also accelerating the commercialization of AI services. Jassy stated that he believes that in the long run, the AI service platform Bedrock, which allows customers to easily use various foundational models, "could become a business comparable in scale to AWS and EC2 (Elastic Compute Cloud)." Furthermore, to help enterprises securely and at scale build and deploy AI agents, Amazon has launched the infrastructure building module AgentCore, whose software development kit (SDK) has been downloaded over 1 million times, with companies like Sony and Ericsson already adopting it
Layoffs Not for Financial Reasons, Aimed at Building the "World's Largest Startup"
Regarding the recent layoffs involving $1.8 billion in severance costs, Jassy provided a strategic explanation. He stated that this decision was not primarily driven by financial reasons or AI replacing human labor, but rather a profound cultural adjustment.
He explained that after years of rapid growth, the company's organizational hierarchy has increased, which may lead to slower decision-making and weaken frontline employees' sense of ownership. To maintain agility and innovation, Amazon's leadership team is committed to operating like "the world's largest startup."
"This means reducing layers, increasing people's autonomy, and acting quickly," Jassy said. He believes that maintaining a streamlined and flat organizational structure is crucial during the current period of technological transformation.
Advertising and Retail Business Steady, AI Empowerment Showing Initial Results
Beyond AWS, Amazon's other businesses are also performing steadily and beginning to benefit from AI technology.
The company's advertising revenue grew by 22% year-over-year this quarter, reaching $17.6 billion, with growth accelerating for three consecutive quarters. Amazon's omnichannel advertising products, combined with its vast user data and precise performance measurement capabilities, continue to attract advertisers. Additionally, by integrating Amazon DSP (Demand-Side Platform) with advertising inventory from streaming services like Netflix, Spotify, and SiriusXM, it further expanded the coverage of its advertising ecosystem.
In the core retail business, AI is also playing a role. Jassy revealed that the AI shopping assistant Rufus has reached 250 million active users this year, and customers using Rufus are 60% more likely to complete a purchase, potentially generating over $10 billion in annual incremental sales. Furthermore, in the fresh grocery sector, Amazon has expanded its same-day delivery service to over 1,000 towns, allowing customers to order fresh and regular items together, which Jassy referred to as a "game changer."
Full Transcript of Amazon's Q3 2025 Earnings Call (translated by AI tools):
Amazon Q3 2025 Earnings Call
Event Date: October 30, 2025
Company Name: Amazon
Source: Amazon
Presentation Segment
Conference Operator:
Thank you for your patience.
Hello everyone, and welcome to Amazon's Q3 2025 financial performance earnings call.
Currently, all participants are in listen-only mode.
After the presentation, we will conduct a Q&A session.
Today's call is being recorded.
In the opening remarks, I will turn the microphone over to Mr. Dave Fildes, Vice President of Investor Relations.
Thank you, sir. Please go ahead.
Dave Fildes, Vice President of Investor Relations:
Hello everyone, and welcome to our Q3 2025 financial performance earnings call
Today, our Chief Executive Officer Andy Jassy and Chief Financial Officer Brian Olsavsky are here to answer your questions.
While listening to today's conference call, we recommend that you refer to our press release, which contains our financial performance, as well as metrics and comments for this quarter.
Please note that unless otherwise stated, all comparisons in this call will be based on our performance for the same period in 2024.
Our comments and responses to your questions reflect management's views as of today, October 30, 2025, and will include forward-looking statements.
Actual results may differ significantly.
Additional information regarding factors that may affect our financial performance is included in today's press release and in the documents we have submitted to the U.S. Securities and Exchange Commission (SEC), including our most recent 10-K annual report and subsequent filings.
During this call, we may discuss certain non-GAAP financial metrics.
You can find additional disclosures regarding these non-GAAP metrics in our press release, the accompanying slides for this webcast, and the documents we have submitted to the SEC (all of which are available on our investor relations website), including reconciliations of these metrics to comparable GAAP metrics.
Our guidance incorporates the order trends we have seen so far and assumptions we believe are appropriate at this time.
Our performance itself is unpredictable and may be significantly affected by many factors, including fluctuations in foreign exchange rates, changes in the global economy and geopolitical conditions, tariffs and trade policies, customer demand and spending (including concerns about economic recession, inflation, interest rates, regional labor market constraints, and the impact of world events), the growth of the internet, online commerce, cloud services, and emerging technologies, as well as various factors detailed in the documents we have submitted to the SEC.
Our guidance assumes, among other things, that we will not engage in any additional business acquisitions, restructurings, or legal settlements.
Accurately predicting the demand for our goods and services is impossible, and therefore our actual results may differ significantly from our guidance.
Now, I will hand the microphone over to Andy.
Andy Jassy, President and CEO:
Thank you, Dave.
We saw strong growth in the business in the third quarter, reporting revenue of $180.2 billion, a year-over-year increase of 12% excluding foreign exchange rate effects.
Operating income was $17.4 billion, but without two special expenses in the third quarter, it would have exceeded $21 billion: one was a $2.5 billion FTC settlement, and the other was an estimated $1.8 billion in severance costs.
Free cash flow for the past 12 months was $14.8 billion.
I will start with AWS.
AWS is growing at a pace not seen since 2022, with year-over-year growth accelerating to 20.2%, the highest growth rate we have seen in 11 quarters. It is worth noting that the year-on-year growth percentage is a relative term.
Achieving a 20% year-on-year growth based on an annualized operating rate of $132 billion is very different from achieving a higher percentage growth rate based on a much smaller annual revenue (which is the case for our competitors).
As of the end of the third quarter, the backlog of orders grew to $200 billion, not including several new undisclosed deals from October, which together exceed our total transaction volume for the entire third quarter.
AWS momentum is strong.
Customers want to run their core workloads and AI workloads on AWS because it offers more powerful capabilities, security, and operational performance.
The scale I see before me gives me confidence in the future.
I will share more details to illustrate why.
It starts with AWS having a broader infrastructure capability.
Startups, enterprises, and governments want to move their production workloads to the place with the broadest and deepest capability set.
AWS has more services and deeper capabilities within those services than any other company and continues to innovate rapidly.
These are the key building blocks for anything customers want to create.
This is also a significant reason why Gartner has named AWS a leader in its Magic Quadrant for Strategic Cloud Platform Services for 15 consecutive years.
We apply the same building block approach to AI.
SageMaker makes it easier for companies to build and deploy their own foundational models.
Bedrock provides customers with leading foundational model options and excellent cost-effectiveness to deploy inference into their next-generation applications.
Most of the value companies will derive from AI in the future will come in the form of agents.
AWS is making significant investments in this area and is well-positioned to become a leader.
Companies will both create their own agents and use agents from other companies.
For those companies building their own agents, the building process is more difficult than it should be.
This is why we launched Strands, which enables builders to create agents more easily using any foundational model they want.
For those companies that have successfully built agents, they have been hesitant to put them into production due to a lack of secure, scalable runtime services, memory, or observability specifically designed for agents.
This is why we launched AgentCore, a set of infrastructure building blocks that allows builders to deploy secure, scalable agents.
Ericsson uses AgentCore to deliver AI agents to all its employees.
Sony built an intelligent AI platform with enterprise-level security, observability, and scalability using it.
Cohere Health is using AgentCore to deploy agents, which will reduce medical review times by up to 30% to 40% The SDK of AgentCore has been downloaded over 1 million times, and our builders are excited about it.
It is an enabler.
The company will also use other people's agents, while AWS continues to build many agents that we believe builders will use in the future.
For coding, we recently opened our intelligent coding IDE called Kiro.
In just the first few days of the preview release, over 100,000 developers flocked to Kiro, and that number has more than doubled since then.
So far, it has processed trillions of tokens, with rapidly growing weekly active users, and developers love its unique specifications and tool invocation capabilities.
For migration and transformation, we offer an agent called Transform.
Year-to-date, customers have used it to save 700,000 hours of manual work, equivalent to 335 developer years of work.
For example, Thomson Reuters uses it to convert 1.5 million lines of code monthly, migrating from Windows to open-source alternatives, and completes tasks four times faster than other migration tools.
Customers have also used Transform to analyze nearly 1 billion lines of mainframe code to migrate mainframe applications to the cloud.
For business customers, we recently launched Quick Suite, bringing consumer-grade AI experiences into the workplace, making it easy to find insights, conduct deep research, automate tasks, visualize data, and take action.
We have seen users shorten projects that would take months to just days, saving over 80% of time on complex tasks and achieving over 90% cost savings.
For contact centers, we offer Amazon Connect, which creates personalized and efficient experiences for contact center agents, managers, and their customers.
Connect has recently surpassed a $1 billion annualized revenue run rate, with 12 billion minutes of customer interactions handled by AI last year, and is being used by large enterprises such as Capital One, Toyota, American Airlines, and Ryanair.
These are real, practical results that customers are achieving, and there are many more similar examples.
Due to its superior capabilities, security, operational performance, and customer-centric approach, AWS continues to win the transformation of most large enterprises and governments to the cloud.
As a result, the vast majority of the company's data and workloads reside on AWS, which is also part of the reason most companies want to run AI on AWS.
To enable customers to do this, we need to have the necessary capacity, and we have been focused on accelerating capacity expansion over the past few months, adding over 3.8 gigawatts of power in the past 12 months, more than any other cloud provider.
To help everyone understand the scale, our current power capacity is double that of AWS in 2022, and we are on track to double it again by 2027. In just the last quarter of this year, we expect to add at least 1 gigawatt of power.
This capacity includes power, data centers, and chips, primarily our custom silicon chip Trainium and NVIDIA chips.
We recently launched Project Rainier, which is our massive AI computing cluster spanning multiple data centers across the United States, containing nearly 500,000 of our Trainium 2 chips.
Anthropic is now using it to build and deploy its industry-leading AI model Claude, and we expect it will use over 1 million Trainium 2 chips by the end of the year.
The adoption rate of Trainium 2 remains strong, fully booked, and has now become a multi-billion dollar business, growing 150% quarter-over-quarter.
Currently, Trainium is being used by a few very large customers, but we expect to accommodate more customers starting with Trainium 3.
We are building Bedrock into the world's largest inference engine, and in the long run, we believe Bedrock could become a business as large as EC2 for AWS.
Most of the token usage in Amazon Bedrock is already running on Trainium.
We are also continuing to work closely with chip partners like NVIDIA, and we continue to place very large orders with them, while also collaborating with AMD and Intel.
These are very important partners, and we expect our relationships to continue to grow over time.
You will see us continue to invest very aggressively in capacity because we see the demand.
The pace at which we are currently increasing capacity is as fast as the pace at which we are monetizing it. (“As fast as we're adding capacity right now, we're monetizing it.”)
This is still at a relatively early stage, representing an unusual opportunity for customers and AWS.
Now I will turn to the store business, where the team continues to deliver and innovate for our key priorities—selection, low prices, and convenience (especially fast delivery)—for customers.
Since last quarter, our selection has increased by 14%, including popular brands like North Face and Charlotte Tilbury, and this year we have added hundreds of thousands of items from popular brands.
Everyday Essentials continues to grow rapidly, with a growth rate nearly double that of other parts of the business year-to-date.
We continue to make it easier for customers to order low-priced perishable food from Amazon, and now customers in over 1,000 towns can purchase fresh food while buying millions of Amazon products, enjoying free same-day delivery
This is a game-changing experience for customers, who can now order milk and electronics simultaneously, check out with a single shopping cart, and have all items delivered to their doorstep within hours.
The team also invented a new "Add to Delivery" button, allowing customers to add items to previously scheduled orders, a feature that has been used over 80 million times since its launch, and it has just been introduced.
This is an example of a seemingly simple yet powerful innovation that makes customers' lives easier.
We remain committed to maintaining a price advantage, matching or beating the prices of other major retailers.
In July, we held the largest Prime Day event ever, where customers saved billions of dollars across more than 35 categories.
We continue to break records in speed.
We expect to achieve the fastest delivery speed ever for global Prime members again this year, and we have begun rolling out a three-hour delivery service in select cities in the United States.
We are also continuing to invest in infrastructure to accelerate delivery speeds in rural areas and serve more customers in more communities.
This includes a commitment to invest over $4 billion in the United States to expand our rural delivery network.
These are small towns where people want fast delivery, but other companies have been pulling back and reducing services.
In contrast, we have increased the number of rural communities that can enjoy our same-day and next-day delivery services by 60%, reaching about half of the total number of communities we plan to expand to by the end of the year.
The store team is also rapidly leveraging AI for innovation.
For example, our AI-driven shopping assistant Rufus has 250 million active customers this year, with monthly users up 140% year-over-year, interactions up 210%, and customers using Rufus during shopping are 60% more likely to complete a purchase.
Rufus is expected to generate over $10 billion in annual incremental sales.
Here are the highlights.
Our generative AI-driven audio feature, which combines product summaries and reviews to make shopping easier, has expanded from hundreds of products at launch to millions of products, with millions of customers using the feature, streaming nearly 3 million minutes.
There’s also Amazon Lens, an AI-driven visual search tool that allows customers to find products using their phone camera, screenshots, or barcodes, now including Lens Live, which can instantly scan products and display real-time matches in a scrollable carousel.
Tens of millions of customers use Amazon Lens each month.
Next is Amazon Advertising.
We are pleased with the continued strong growth, generating $17.6 billion in revenue this quarter, a 22% year-over-year increase.
We see strong performance from our comprehensive full-funnel advertising product portfolio, helping advertisers reach an average of over 300 million ad-supported audiences in the United States alone
We are also excited about our demand-side platform Amazon DSP, which allows advertisers to plan, activate, and measure full-funnel investments.
Last quarter, I mentioned our partnership with Roku, and building on that, we established a partnership with Netflix to provide advertisers using Amazon DSP direct access to Netflix's premium advertising inventory.
We also announced integrations with Spotify and SiriusXM.
Through Spotify, we provide advertisers with programmatic direct access to over 400 million monthly active ad-supported listeners worldwide.
Through SiriusXM, brands can reach 160 million monthly active digital listeners on services like Pandora and SoundCloud.
We are excited about the advertising opportunities brought by live sports programming on Prime Video.
Live sports programming has generated significant interest from advertisers in early negotiations for 2025-26, and we have exceeded our expectations in early commitments, achieving substantial growth across the board.
Finally, we continue to innovate for advertisers using AI.
For example, in September, we announced a smart AI tool in Creative Studio that can plan and execute the entire creative process in hours instead of weeks.
We are also innovating in several other areas and seeing strong momentum, and I will mention a few of them.
In Prime Video live sports programming, NBA games on Prime kicked off last week, and our opening night doubleheader attracted an average of 1.25 million viewers in the U.S., a double-digit increase over last season's cable ratings.
You will see us bring the same continuous innovation here as we do in our NFL broadcasts.
We will add the Masters Tournament in 2026 and launch a new hole-by-hole tournament in partnership with the PGA Tour this Black Friday.
We have added Peacock and Fox One to the additional subscription services of over 100 channels on Prime Video in the U.S.
We are thrilled with the response to Alexa+.
Compared to what we call the classic Alexa experience, Alexa+ customers interact with Alexa twice as much.
These interactions are longer and cover a wider range of topics.
They use Alexa+ on Fire TV 2.5 times more frequently than the classic version, discover audio content through natural conversation four times more often, browse photos four times more, and complete shopping conversations that end in purchases four times more.
We have expanded the number of Project Kuiper satellites in space to over 150 and achieved speeds exceeding 1 Gbps in testing using our enterprise-grade customer terminals, marking the first known commercial phased array to break that threshold
Finally, the Zoox robot taxi has been opened to passengers in Las Vegas, and we have announced Washington D.C. as the eighth testing location.
We look forward to these services continuing to be promoted to more passengers.
The fourth quarter is one of the busiest and most exciting times of the year for us, and we are excited about the continued demand for AWS, the innovations to be announced at the reInvent conference in December, the positive response from customers to our AI-driven experiences, all the gifts we will deliver throughout the holiday season, and more.
I would like to thank our colleagues around the world in advance, who are preparing to serve customers again.
With that, I will hand the microphone over to Brian for a financial update.
Brian T. Olsavsky, Senior Vice President and Chief Financial Officer:
Thank you, Andy.
Starting with our revenue performance, global revenue was $180.2 billion, an increase of 12% year-over-year, excluding a favorable impact of 90 basis points from foreign exchange.
In the third quarter, we reported global operating income of $17.4 billion.
This operating income includes two special expenses that reduced operating income by $4.3 billion.
The first expense of $2.5 billion relates to a legal settlement with the Federal Trade Commission, affecting the North America segment, and is recorded in other operating expense items.
The second expense of $1.8 billion relates to severance costs from role eliminations, affecting all three of our segments.
Severance costs are primarily recorded in the technology infrastructure, sales and marketing, and general and administrative expense items.
Excluding these two expenses, global operating income would have been $21.7 billion, which is $1.2 billion above the high end of our guidance range.
Turning to our segment performance.
We are encouraged by the innovations our teams are bringing to customers across all three segments.
In the North America segment, third-quarter revenue was $106.3 billion, an increase of 11% year-over-year.
International segment revenue was $40.9 billion, growing 10% year-over-year when excluding foreign exchange impacts.
The number of global paid units increased by 11% year-over-year.
We continue to prioritize investments that matter most to customers.
In the third quarter, our sharp pricing, broad selection, and fast delivery times continued to resonate with customers.
Customers appreciate being able to quickly receive the items they need daily, including perishable food, delivered on the same day.
Our millions of third-party sellers globally continue to make significant contributions to our vast selection, helping customers find the items they need at competitive prices.
We are committed to building innovative services and features for sellers, including our ongoing advancements in generative AI.
Today, over 1.3 million sellers are using our generative AI capabilities to publish high-quality product information more quickly
Better product information translates into better customer attraction. In the third quarter, the share of global third-party seller units was 62%, an increase of 200 basis points compared to the third quarter of last year.
Turning to profitability, the operating profit of the North America segment was $4.8 billion, with an operating profit margin of 4.5%.
Excluding the $2.5 billion expense related to the FTC legal settlement, the operating profit of the North America segment would have been $7.3 billion, with an operating profit margin of 6.9%.
The operating profit margin of the North America segment also includes the impact of some severance costs.
The operating profit of the international segment was $1.2 billion, with an operating profit margin of 2.9%.
Excluding the impact of severance costs, the operating profit margin of the international segment has expanded year-on-year.
Globally, our progress in key inputs is providing a better customer experience while driving a more efficient cost structure.
For example, we have made significant progress in improving inventory layout to accelerate delivery to customers.
Therefore, for the third consecutive year, we expect to achieve the fastest delivery speed ever for Prime members in 2025.
We continue to adjust and improve our fulfillment operations, and our regional network is scaling up.
We have seen many benefits from improvements in the inbound process, including a reduction of nearly four days in the U.S. inbound lead time compared to last year.
This allows us to procure inventory more efficiently, benefiting working capital.
We are also placing inventory more strategically throughout the network.
By leveraging our existing infrastructure, we now offer U.S. customers the ability to order perishable food and have it delivered the same day within just five hours.
We have seen positive early results.
Since the launch in January, when customers began purchasing fresh food on Amazon, they have visited the site more frequently, and their repeat purchase rate is twice that of non-fresh food shoppers.
Looking ahead, we see further opportunities to improve productivity in our global fulfillment and transportation network.
We will continue to improve inventory layout to reduce transportation distances and the number of touches per package.
We will also continue to evolve based on the benefits brought by regional networks through algorithm improvements and the introduction of robotics and automation.
While operating profit margins may fluctuate quarterly, we have a thoughtful approach to achieving sustained long-term progress.
Turning to the advertising business, advertising revenue was $17.7 billion, and growth accelerated for the third consecutive quarter.
We continue to see strong growth on an increasingly large base, as our full-funnel advertising approach connecting brands with customers resonates.
Next is our AWS segment, with revenue of $33 billion, a year-on-year increase of 20.2%.
This accelerated by 270 basis points compared to the previous quarter, primarily driven by strong growth in our AI and core services, as well as more capacity brought online to support customer demand.
AWS revenue increased by $2.1 billion quarter-on-quarter, now achieving an annualized revenue run rate of $132 billion
AWS operating profit was $11.4 billion, reflecting our continued growth and our efforts to drive efficiency across the business.
We are expanding our data center footprint primarily to accommodate generative AI, and the related depreciation will indeed impact our margins as these assets come online.
As we have long stated, we expect AWS operating margins to fluctuate over time, partly due to the level of investment we are making at any given point.
Now let's talk about our cash capital expenditures (Cash CapEx), which were $34.2 billion in the third quarter.
Year-to-date, we have spent $89.9 billion.
This is primarily related to AWS, as we invest to support the demand for AI and core services, as well as custom silicon like Trainium, while also including the technology infrastructure that supports our North American and international operations.
We will continue to make significant investments, particularly in AI, as we believe this is a huge opportunity with strong long-term capital return potential.
Additionally, we continue to invest in our fulfillment and transportation network to support business growth, improve delivery speed, and reduce service costs.
These investments will support growth for many years to come.
Looking ahead, we expect full-year cash capital expenditures of approximately $125 billion in 2025, and we anticipate this amount will increase in 2026.
I will conclude my remarks with net income.
While we primarily focused our comments on operating profit, our net income for the third quarter was $21.2 billion, which includes $9.5 billion in pre-tax earnings related to our investment in Anthropic.
This investment activity is unrelated to Amazon's ongoing operations and is included in non-operating profit.
We are encouraged by the start of the holiday season and are ready to serve customers in the coming months.
I want to thank the teams at Amazon around the world for their hard work as we prepare to delight customers this holiday season.
We are committed to enhancing the customer experience, which is the only reliable way to drive sustainable value for shareholders.
With that, let's move into the Q&A session.
Q&A Session
Conference Operator:
Q&A Session
Conference Operator:
Thank you.
We will now open the call for questions. (Operator Instructions)
Our first question comes from Justin Post of Bank of America. Please go ahead.
Justin Post:
Okay. Thank you.
I want to ask about AWS.
Can you talk about your feelings on the current capacity levels and how constrained capacity is at the moment?
Then in your prepared remarks, you mentioned the demand for Trainium 3 and the potential to expand the customer base
Can you talk about the demand for Trainium that you see outside of your major customers? Thank you.
Unspecified speaker:
Sure.
Regarding capacity, as I mentioned in my opening remarks, we have added quite a bit of capacity.
We added 3.8 gigawatts of capacity over the past year, and we will add more than 1 gigawatt in the fourth quarter.
We expect to double our total capacity by the end of 2027.
So we are bringing in quite a bit of capacity.
Currently, in the entire industry, the bottleneck may be power.
I think at some point it may shift to chips, but we are bringing in a lot of capacity.
The pace at which we are bringing in capacity is as fast as the pace at which we are monetizing it.
Then there is the demand for Trainium outside of major customers.
First, as I mentioned, Trainium 2 is performing very well.
Trainium 2 is fully booked.
It is now a business worth billions of dollars.
Its revenue has grown by 150% quarter-over-quarter.
And you can now see truly large-scale projects, such as Project Rainier, which we are collaborating on with Anthropic, where they are training the next version of Claude on 500,000 Trainium 2 chips, and by the end of the year, they will use 1 million Trainium 2 chips.
As I mentioned, currently for Trainium 2, we only have a few very large customers using it.
But because Trainium has a 30% to 40% cost-performance advantage over other available options, and because customers are considering scaling their production workloads more broadly, shifting to AI-focused work, and using inference, they are very concerned about cost-performance.
So we have a lot of demand for Trainium.
Trainium 3 should be previewed by the end of this year, with more substantial supply starting in early 2026.
We have many customers, including very large ones, as well as what I call medium-sized customers, who are quite interested in Trainium 3.
Conference operator:
The next question comes from Brian Nowak of Morgan Stanley.
Brian Nowak:
Thank you for answering my question. Congratulations on this quarter's performance, everyone.
Perhaps I have two questions.
First, Andy, a somewhat philosophical chip question.
There are many questions in the market about Trainium and its positioning relative to other third-party chips.
So, what key obstacles do you think Trainium 3 needs to overcome to truly expand adoption, as you mentioned in the previous question, and continue to drive the development of Trainium rather than relying on third-party chips to meet potentially broader demand in the near term?
Andy Jassy, President and CEO:
Okay.
First of all, we always provide our customers with a variety of chip options.
This has always been the case in every major technology building block or component of AWS.
In fact, throughout AWS's history, no player has ever dominated an entire segment for a long time and met everyone's needs across all dimensions.
Therefore, we have a very deep partnership with NVIDIA.
This has been the case for a long time and will continue to be so for the foreseeable future.
We purchase a large number of NVIDIA chips.
We have no restrictions on purchasing NVIDIA.
I expect we will continue to purchase more NVIDIA chips next year and in the future.
But unlike most tech companies, we have our own very strong chip team.
That is our Annapurna team.
You first saw the results of our collaboration with Graviton in CPUs, which offers about a 40% better cost-performance ratio compared to other x86 processors.
You now see the results of Trainium in AI custom silicon, which provides customers with a roughly equivalent cost-performance advantage compared to other GPU options.
And our customers, in order to use AI as broadly as they wish, and remembering that we are still in a relatively early stage, will need better cost-performance, and they care a lot about this.
So, I previously mentioned the momentum of Trainium 2.
I believe for us, when we consider Trainium 3, I expect Trainium 3 will be about 40% better than Trainium 2.
And Trainium 2 already has a significant advantage in cost-performance.
So, we certainly have to deliver this chip.
We must deliver it in large quantities and quickly.
And we must continue to invest in the software ecosystem, which has been getting better.
As we have more proof points like Project Rainier, and what Anthropic has done on Trainium 2, it builds increasing credibility for Trainium.
I think customers are very optimistic about this.
I am equally optimistic.
Conference Operator:
Our next question comes from Doug Anmuth of JP Morgan. Please go ahead.
Doug Anmuth:
Thank you for taking my question.
I'm basically still on the same topic, Andy
But can you talk about the architecture of Project Rainier, what differentiates it, and what it means for customers and AWS?
Do you expect Rainier to expand beyond Anthropic?
And how do you replicate Rainier with the Trainium 3 chip? Thank you.
Andy Jassy, President and CEO:
Okay, I think what makes Project Rainier attractive to Anthropic is that it is really based on the Trainium 2 chip, and we have built a very large cluster that they can use in a very broad way.
And building a cluster with over 500,000 chips and soon reaching 1 million is not simple.
This is an infrastructure feat that is difficult to achieve at scale.
So part of it is our long-term infrastructure capabilities that we have built in AWS, which are unusual in the industry.
But it is also just the performance and cost-effectiveness of the chips, both of which are important.
And I believe Project Rainier is specifically designed for Anthropic.
But we also have many other customers interested in using large Trainium chip clusters, and we hope to give them the opportunity to do so with Trainium 3.
Conference Operator:
The next question comes from Mark Mahaney of Evercore ISI. Please go ahead.
Mark Mahaney:
Thank you.
I want to ask about two topics, groceries and how you view future employee numbers.
Regarding groceries, I think—perishables, I believe last quarter you mentioned that about 70% of users had never purchased perishables from Amazon before.
Please talk about whether you—I think you previously used the term "game changer."
Does this mean that perhaps Amazon Fresh stores are no longer needed?
You have historically had this density advantage with DVD delivery trucks.
Do you think you have reached a critical mass and speed that can really change people's habits and make them truly view Amazon as one of their preferred grocery options?
Do you really think we have reached that point?
And secondly, regarding employee numbers, given some recent news, please talk to us about how you view future employee numbers.
Do you see—are the efficiency levels gained from AI sufficient to keep employee numbers relatively stable in the foreseeable future?
Please discuss the pros and cons or gains and losses regarding employee numbers in this regard. Thank you.
Andy Jassy, President and CEO:
Okay.
I'll start with groceries, Mark.
We have a very large grocery business.
If you look at our entire grocery business, even without counting Whole Foods Market and Fresh, the total gross merchandise sales (GMS) have exceeded $100 billion in the past 12 months, which would make us one of the top three grocery retailers in the United States.
A large portion of this consists of many products you find in the middle aisles, such as consumer goods, canned foods, pet foods, and health and beauty products.
Very important.
This part of the business continues to grow at a very good pace.
But we also have Whole Foods Market, which is a pioneer in organic foods, and its growth rate is faster than most grocery retailers, with a good profitability trajectory.
We will expand the physical store footprint of Whole Foods in the coming years.
I am also very excited about our new daily shop concept, which is a smaller version of Whole Foods in an urban environment, and we have opened three of them.
They are off to a great start, and you should see more.
Now, as you mentioned, we have been considering having a larger physical store presence.
We continue to experiment with various formats.
But what we are most excited about is what you mentioned, which is the ability to offer perishable foods and achieve same-day delivery.
If you think about how many customers purchase from us multiple times a week, buying items like shampoo, detergent, paper cups, or water, then being able to add milk, eggs, yogurt, and other perishables to their orders and have them in the same shopping cart, delivered a few hours later, is very appealing.
We started this about a year ago in several markets, and we are very surprised by the adoption rate, not only in how quickly people started buying perishables from us, but also in the frequency with which they returned to purchase perishables and groceries.
So, we have now expanded it to 1,000 cities in the U.S.
By the end of this year, it will reach 2,300 cities.
This is indeed changing the trajectory and scale of our grocery business.
I also believe that the traditional weekly stock-up grocery shopping habit that has lasted for years is changing.
I think we are an important driver of this, and I believe there is huge potential in this area.
This does not mean we will not continue to experiment with other physical formats, but we have indeed found a very important direction in leveraging same-day delivery facilities for perishables.
Then regarding your personnel question, what I want to tell you is that the news we announced a few days ago was not really driven by finance, nor was it driven by AI.
At least not right now.
It is actually a cultural issue.
If you have experienced rapid growth like us for several years, the scale of the business, the number of personnel, the number of locations, and the types of businesses engaged will ultimately lead to having significantly more personnel and more levels than in the past.
When this happens, sometimes unknowingly, you may weaken the sense of ownership of those who are doing the actual work and have most of the "two-way door" decision-making authority (decisions that should be made quickly at the front line), which can lead to a slowdown in speed.
As a leadership team, we are committed to operating like the largest startup in the world.
This means reducing levels, increasing ownership among personnel, and enabling rapid invention and action.
In Amazon's history, or perhaps in business history, there has never been a moment like now, where it is so important to remain streamlined, flat, and act quickly as technology transformation occurs, and that is exactly what we aim to do.
Meeting Operator:
The next question comes from Eric Sheridan of Goldman Sachs. Please go ahead.
Eric Sheridan:
Thank you very much for taking my question.
Andy, could you reflect on the ongoing opportunities presented by the rollout of more robotics and automation, as well as the broader theme of physical AI in operations?
How should I think about this as both a driver of potential efficiencies and a driver of the ability to reinvest back into the business over the long term? Thank you very much.
Andy Jassy, President and CEO:
Robotics is a very important area of investment for us.
Currently, we have over 1 million robots in our fulfillment network.
What I would say is that while this is substantial, we have many more inventions in progress.
So I expect that over time, we will have more robots.
Robotics is very important to us, our customers, and our colleagues because it enhances safety, boosts productivity, increases speed, and allows our human colleagues to focus on problem-solving and what they do best.
We expect our personnel to continue to be the core and center of our fulfillment network, just as they were when we first began exploring robotics.
We anticipate that over time, we will have a fulfillment network where robots complement and collaborate with humans.
But I think you will continue to see us invest significantly in robotics.
This will help improve safety, productivity, speed, and ultimately lower some costs, allowing us to continuously enhance the customer experience.
Meeting Operator:
The next question comes from John Blackledge of TD Cowen. Please go ahead
John Blackledge:
Okay. Thank you.
How does Amazon view the future of agentic commerce?
How do you think Amazon will leverage agents to serve customers and enable them to purchase products on Amazon? Thank you.
Andy Jassy, President and CEO:
I am personally very excited, and as a company, we are very excited about the long-term prospects of agentic commerce.
It has the potential to benefit customers.
It has the potential to be very beneficial for e-commerce.
I think if you want to buy something, there are few experiences better than coming to Amazon.
But if you don’t know what you want, then having physical salespeople in a physical store still has some advantages.
Clearly, many people have been doing this on Amazon, but you often want to ask questions and seek help to narrow down what you are looking for.
When you keep asking new questions, you hope to see a bunch of different options presented to you.
I believe AI and agentic commerce will change the online experience, making the experience of filtering when you don’t know what you want better than in a physical environment.
Now, we clearly have our own efforts in this area.
We have Rufus, which I mentioned in the opening remarks, and it is continually getting better and being used more widely.
We also have features like "Buy For Me," where we showcase products on Amazon that we don’t carry ourselves but are available from other sellers.
Then if customers want us to purchase from those sellers' websites on their behalf, we will do so.
Both of these features have been successful for us.
But we are also in conversations with third-party agents and expect to collaborate with them over time.
I think this somewhat reminds me of when search engines first started years ago; they were a source of commercial discovery.
You have to find the right way to collaborate.
Today, search engines account for only a small portion of our referral traffic.
And third-party agents are just a very small subset of that.
But I do believe we will find ways to collaborate.
However, we need to find a way to improve the customer experience.
Right now, I think the customer experience is not good.
There is no personalization.
There is no shopping history.
Delivery estimates are often inaccurate.
Prices are often incorrect as well.
So we need to find a way to improve the customer experience and make the right value exchange.
But I think the exciting part and the prospect is that AI and agentic commerce solutions will scale online shopping I believe this is very beneficial for customers.
I also think this is very beneficial for Amazon.
Because ultimately, you will buy from the company that offers you the widest selection, great value, and consistently fast and reliable service.
I think this is a positive for us.
Conference Operator:
Our last question comes from Colin Sebastian at Baird. Please go ahead.
Colin Sebastian:
Thank you. Good afternoon.
First, regarding AWS, following up on the previous topic, how much of this acceleration is driven by core infrastructure and how much is driven by the monetization of AI workloads?
I want to understand how important new services like AgentCore are in attracting enterprises to use AWS to build agents.
And second, regarding the acceleration of the advertising business, if possible, could you differentiate between the core advertising contribution and the contributions from DSP and Prime Video? That would be very helpful. Thank you.
Andy Jassy, President and CEO:
Okay, I'll start with AWS.
We are very pleased with our performance this quarter.
Achieving 20% year-over-year growth on an annualized operating rate of $132 billion is unusual.
We have strong momentum.
You can see that.
We are seeing acceleration in growth in both the AI space and core areas.
We are seeing growth in AI, including inference, training, and the use of our Trainium custom silicon.
Bedrock continues to grow rapidly.
SageMaker is also growing quickly.
I believe the number of companies building agents is quite substantial.
I truly believe that most of the long-term value companies will derive from AI will come from agents.
And I think that building agents is still harder than it should be right now.
You need tools to make it easier, which is why we are building Strands, an open-source capability that allows people to build agents with any model they can imagine.
But more importantly, when you talk to enterprises or companies that care deeply about security and scalability, they start building agents, but they don’t feel they have the building blocks to create the kind of secure, scalable agents that their business, customer experience, and data rely on.
That’s the inspiration behind AgentCore—building another set of foundational building blocks, just like we built compute, storage, and databases in the early days of AWS.
We defined a set of building blocks you need to be able to deploy agents securely and at scale, and we provide those in AgentCore
When we talk to customers, it resonates strongly.
There is nothing else like it.
It is changing their timelines and their acceptance of building intelligent agents.
This is very attractive to them.
So, I think the work we are doing to enable the secure, scalable building and operation of intelligent agents, along with some of the intelligent agents we are building ourselves that excite customers, is an attractive combination for them.
Yes.
I think another growth point we see at AWS is that many enterprises have resumed migrating their on-premises infrastructure to the cloud.
We continue to win a significant share of these transformations.
Looking at our current momentum, I believe we can continue to grow at this pace for some time.
I think in advertising, this is also an area where we collectively feel very satisfied with the progress.
All of our advertising products grew meaningfully this quarter.
I think several things are happening.
We have what I believe is a rather unusual full-funnel product.
If you look at the top of the funnel, it is usually about building brand awareness and broad reach, leveraging our own Prime Video and live sports capabilities, all the way down to point-of-sale using sponsored products.
Most companies do not have such a strong full-funnel product as we do.
And then combined with the audience curation and development we can do, along with advantage measurement, all of this leads to a very unusual return on advertising spend (ROAS).
I believe we have multiple areas where we can expect to continue growing.
One is in our store business; I still believe that if you look at global retail market share, 80% to 85% is still in physical stores, and this equation will flip over time.
I think AI will only accelerate this process.
So I believe we still have huge opportunities in existing stores.
Then there is video; we have not been in this space for long, but it has already become a very large source of advertising revenue.
I think we are still in a relatively early stage, and I believe this will continue to be a huge growth area.
And as you mentioned, the demand-side platform or Amazon DSP is also growing rapidly.
Part of the reason relates to some of the capabilities we have.
We have always had some core components that people want, around some of our properties, measurement capabilities, Amazon Marketing Cloud.
But for a while, during our DSP build-out, we were missing some features that customers told us were important to them.
Over the past 20 months, the team has significantly closed those gaps to the point where people now feel our DSP is fully featured.
Then you look at some of the partnerships we have built
Our partnership with Roku has given us the largest connected TV coverage in the United States.
Coupled with what we've recently done to provide our DSP clients with the opportunity to integrate advertising inventory from Netflix, Spotify, and SiriusXM, this is very powerful.
Therefore, our demand-side platform is growing very rapidly.
As a result, we are very optimistic about the work being done there.
Clearly, we still have work to do, but I don't think we are close to the limits of growth.
Dave Fildes, Vice President of Investor Relations:
Thank you for joining our call today and for your questions.
The recording will be available on our investor relations website for at least three months.
We appreciate your interest in Amazon and look forward to speaking with you again next quarter.
The event has ended
Disclaimer:
This transcript may not be 100% accurate and may contain spelling errors and other inaccuracies
