Amazon Conference Call: Cash Flow 'Burned Through', but AI Is a 'Once-in-a-Lifetime Opportunity'; AWS Targets Trillion-Dollar Revenue, Retail Business Undervalued

Driven by strong demand for AI computing power, Amazon's AWS cloud business achieved its fastest growth rate in nearly five years in Q3, with management unusually providing forward guidance that "AWS will grow by over 40% for the next two years and is on the path to $1 trillion in revenue."

Wallstreetcn mentioned that after the US market closed on October 29, Amazon disclosed that its Q3 revenue increased by 12% year-over-year, slightly below analysts' expectations, and EPS declined by 3% year-over-year instead of rising. Operating profit margin approached 14%, a record high.

The financial report's biggest highlight remains the AWS cloud business. Q3 AWS net sales climbed 39.3% year-over-year to $45.98 billion, beating analyst forecasts by 1.3% and marking the fastest single-quarter growth in nearly five years. The annualized revenue run rate at quarter-end reached $184 billion, up 8.9% from the end of Q2.

During the earnings call, Amazon CEO Andy Jassy described the current AWS growth logic as a self-reinforcing "flywheel": Leading features and security attract customers to migrate to the cloud; centralized data and applications on the cloud in turn drive AI inference demand; AI inference boosts core computing, storage, and network consumption, thus driving continuous reinvestment in AWS's foundational capabilities.

Jassy expects AWS to grow by over 40% in both 2027 and 2028, with the possibility of annual revenue surpassing $1 trillion in the next five to ten years. Meanwhile, the company's backlog orders have reached $578 billion, showing triple-digit growth year-over-year.

Management also emphasized that AI demand will continue to drive up capital expenditure. Jassy described AI as a "once-in-a-lifetime opportunity" and stated that the company has the capacity to fund these plans. Amazon expects cash capital expenditures of around $220 billion in 2026, most of which will go to AWS, and anticipates a "significant increase" in 2027.

As Amazon's profit engine, AWS's performance this quarter eased market concerns about "scale being too large to sustain high growth."

AWS Q3 revenue reached $46 billion, up 39.3% year-over-year, the sixth consecutive quarter of accelerating growth and the fastest pace in nearly five years. The current annualized revenue run rate is as high as $184 billion.

CEO Andy Jassy described the feverish demand to investors using highly visual language at the meeting:

We are installing an AI accelerator roughly every 14 seconds, and almost as soon as we get new capacity online, customers snap it up.

It's not just the current boom—future order pools are equally bottomless. According to disclosures, AWS's unfulfilled backlog now stands at a staggering $578 billion, a triple-digit increase year-over-year.

Looking ahead, Jassy provided highly imaginative performance guidance:

On such a large base, we expect AWS to grow by over 40% annually in both 2027 and 2028. We believe AWS will become a trillion-dollar business in the next 5 to 10 years. This is truly an extraordinarily rare opportunity.

Amazon has listed Bedrock as a key service for AWS AI growth.

The company said Fortune 500 spending on Bedrock is up nearly 7x year-over-year, and total tokens processed by Bedrock are up nearly 17x, making Bedrock the fastest-growing multibillion-dollar service in AWS history.

Jassy positions Bedrock as a multi-model aggregation platform, and believes there will be six to eight leading models coexisting within the next two to three years, with most enterprises using multiple models and even blending them in the same generative AI applications.

He likened Bedrock's long-term potential to EC2, saying its scale "may be as large as EC2."

The market has always closely watched whether Big Tech's enormous capital inputs in AI can translate into actual profits.

In Q3, Amazon's capital expenditure reached $75.9 billion, mainly for AWS and AI infrastructure to support strong customer demand. Management is confident AI is a massive opportunity, expected to drive long-term growth in both revenue and free cash flow, as well as superior returns on invested capital.

For the twelve months ending September 30, Amazon's operating cash flow reached $168.3 billion, up 29% year-over-year. However, free cash flow was negative $42.4 billion, compared to positive $14.8 billion in the same period last year.

Amazon maintains its expectation for around $220 billion in cash capital expenditure for the full year, with no upward revision in this report, but anticipates a "significant increase" in 2027 spending.

In the face of this "burn rate," Jassy's stance is unusually firm. He candidly admits this is a strategy for building a long-term platform, not a short-term data center rental:

AI is our once-in-a-lifetime opportunity; it is the greatest technological shift of our lifetime. If we can keep a $184 billion annual-revenue business growing over 40% for two years and ultimately move towards a trillion dollars, to maintain our leadership position, we must invest.

Jassy summed up the significance of this bet in one sentence:

We have the capacity to fund this plan, and I believe it will forever change the scale of Amazon, forever change our revenue and free cash flow.

Beyond buying compute, the benefits of Amazon's in-house chips are showing.

In Q3 alone, customers using Amazon's custom chips (Graviton and Trainium) saved over $2 billion. Commitments for Trainium and Graviton are up more than 20x year-over-year. Jassy revealed:

Due to such strong demand for Trainium, customers have been urging us to offer it outside AWS, and we intend to do so by the end of 2027.

In e-commerce, Q3 sales rose only 2% year-over-year, marking the slowest quarterly growth since early 2023. Amazon's guidance midpoint for holiday season and Q4 sales also fell short of analysts' estimates, suggesting online business may not rebound quickly in the short term.

However, management emphasized that, excluding the impact of Prime Day timing differences, North America and international business still grew by 6% and 8%, respectively.

This year, Amazon rescheduled its Prime Day event: in most major markets, including the US, Prime Day shifted to Q2, while 2025's Prime Day is set entirely for Q3.

Excluding the impact of Prime Day timing, Q3 2026 year-over-year growth would be about 4 percentage points higher. Management believes customer engagement in Q3 was strong, maintaining growth momentum from H1. Jassy especially reminded the market not to underestimate the progress in its physical stores and retail business:

I think some people have somewhat overlooked how much progress this business has made. I don't know of any other retailer with revenue anywhere near Amazon's hundreds of billions scale, maintaining double-digit operating margins (including advertising), and still growing at this speed.

Among these, fresh groceries and daily necessities have become major traffic drivers, with growth three times that of other retail categories.

Data shows that customers who purchase fresh food and use same-day delivery buy seven times more daily necessities than those who don't use the service; likewise, these fresh food customers also spend over twice as much on non-grocery categories (such as home and apparel) compared to other customers.

In terms of cost reduction and efficiency, robotics technology deserves much credit. CFO Brian Olsavsky noted that there are now over 1 million robots deployed in Amazon's fulfillment network.

In 2026, Amazon doubled the fleet size of its robotic arms (such as Cardinal and Sparrow), enabling the company to reduce per-unit fulfillment cost for the fourth year in a row despite headwinds from rising fuel and shipping costs.

About the advertising business, Q3 advertising revenue hit $19.6 billion, with slowed growth mainly due to Prime Day schedule adjustments and other short-term factors, dragging the growth rate by about 8.5 percentage points. Management expects Q4 ad revenue growth to rebound sharply to around 20%.

For the upcoming Q4, Amazon gave strong guidance: net sales expected between $240 billion and $247 billion; operating profit forecast between $28.5 and $33.5 billion.

Full Q3 earnings call transcript (AI-assisted translation):

Welcome to Amazon's 2026 Q3 Financial Results Conference Call. All participants are in listen-only mode. There will be a Q&A session after management's remarks. This call is being recorded.

Remarks by Vice President of Investor Relations, Dave Fildes

Welcome to our Q3 2026 financial results call. On today's call to address questions are CEO Andy Jassy and CFO Brian Olsavsky.

For those listening in, we suggest referencing our press release, which includes the quarter's financial results, metrics and related notes. Unless otherwise stated, all comparisons in this call refer to performance in the same period of 2025. Today's comments and Q&A represent management's views as of October 29, 2026, and include forward-looking statements, which may materially differ from actual outcomes.

Potential factors affecting our financial results are detailed in today's press release and in our filings with the SEC, including the latest annual report (Form 10-K) and subsequent filings. During this call, we may discuss certain non-GAAP measures. Related disclosures and reconciliations to GAAP measures are available on our Investor Relations website.

Our guidance reflects order trends observed to date and what we believe are reasonable assumptions, but actual results are inherently unpredictable and may be materially affected by numerous factors, including: exchange rates and energy prices, changes in global economic and geopolitical conditions, tariffs and trade policy, supply volatility (including memory chips), demand and consumer spending changes (including recession concerns, inflation, interest rates, and regional labor constraints), global events, growth in internet/online commerce, trends in cloud services and emerging technologies, and other factors detailed in our SEC filings.

Our guidance assumes no additional acquisitions, restructurings, or legal settlements. As it's difficult to accurately predict market demand for our products and services, actual results may differ materially from guidance.

With that, I’ll turn the call over to Andy Jassy.

CEO Andy Jassy Remarks

Revenue and Profit Overview

This quarter's revenue was $201.9 billion, up 12% year-over-year; operating profit was $28.1 billion, up 61% year-over-year. Q3 was another strong quarter for Amazon.

AWS Business

AWS's strategy is executing effectively. Despite a high comparison base due to significant expansion last year, AWS still achieved year-over-year growth of 39.3%, accelerating for the sixth straight quarter, marking its fastest growth in nearly five years. Contracted backlog reached $578 billion, with triple-digit year-over-year growth and diversified structure. We are installing an AI accelerator every 14 seconds, and almost as soon as we add new capacity, customers consume it. Revenue for core compute and storage both saw their fastest growth in years. Core business operating margins remain high over time; margins for our AI business are also rapidly improving, well ahead of where the core business was at a comparable stage—even with rising component costs and substantial depreciation pressures from capex. We're accelerating capex payback cycles, and more convinced than ever: our AWS investments will create a business far larger than we envisioned two years ago, generating higher revenue, free cash flow, and return on invested capital.

AWS is developing a powerful flywheel effect—borrowing Jim Collins’ "flywheel" concept from "Good to Great": Leading features, security, and operational performance drive customer application and data migration to AWS; that migration drives AI demand, since customers want inference close to their data/applications; AI growth boosts core AWS demand, as AI relies on core services for reinforcement learning, post-training, and tool calling; all this leads to ongoing reinvestment in AWS capabilities. This is nothing like simple GPU hosting or short-term data center rent—it’s a long-term platform strategy—helping customers broadly innovate and iterate user experiences, operate reliably, and build durable partnerships for years. This strategy also generates attractive long-term economics, not reliant on supply scarcity.

AWS Progress Examples

For the core business, it's notable that ~85% of global IT spend still happens on-premises. Over the next 10–20 years, that will shift to the cloud. Industry once thought critical workloads like mainframes or SAN storage would never move to the cloud, but with AWS launching unique cloud alternatives and AI-driven migration, these workloads are accelerating to the cloud. Enterprise customers benefit: lower capex, reduced operating costs, faster innovation, higher developer productivity, better data access, and easier AI adoption. Our core business enjoys attractive margins, partly thanks to deep operational experience from 20 years of diligence. As the core business grows, we continue reinvesting resources into AI feature development.

A case in point: As customer data volumes boom and firms seek value via analytics/AI, moving data between storage and analytics engines can be cumbersome/expensive. By acquiring DuckLabs, DuckDB will become AWS’s universal SQL dialiect for analytics, modernizing data plumbing—fast, easy, and broad coverage. Customers can write one SQL query, then execute it on any AWS analytics service. This is an industry-first and a breakthrough for enabling next-generation agent queries without costly migrations.

On the AI side, we’re constantly expanding Amazon Bedrock inference capabilities, driving remarkable growth: Fortune 500 spend on Bedrock is up nearly 7x, and Bedrock token volume is up nearly 17x year-over-year. Bedrock is the fastest-growing multibillion-dollar service in AWS history. Recently, we added OpenAI models to Bedrock (usage quickly ramping up), plus new Bedrock-hosted models from Kimi, GLM, and SpaceX.

Security is now inseparable from any conversation on AI—Bedrock's secure-by-design setup is a top reason customers choose it. Data privacy is guaranteed: Input/output is never shared with model providers or used to train base models; when customers fine-tune, they get a private model instance solely for their use, and all transferred/stored data is encrypted.

As cutting-edge model capabilities grow, they find previously undetected software vulnerabilities at unprecedented speed. For this, we created AWS Continuum, which discovers, prioritizes, verifies, and patches code flaws at machine speed. Capital One, IBM, Infor, Luminate, and Robinhood already use it. In Q3, we announced Continuum integration into code-writing tools, including Anthropic’s Claude Code, OpenAI’s Codex, and AWS’s Cura. Customer interest exceeded any prior security service launch by a wide margin.

Agent deployment keeps accelerating, many built on Bedrock Agent Core. There are now tens of thousands of Agent Core customers; agents deployed in Q3 doubled the tally from the previous three quarters combined. Besides building dedicated agents, most enterprises use out-of-the-box agent services; we offer popular products: Kiro for programming, Amazon Connect for contact centers, Amazon Transform for migrations, and Amazon Q as a smart AI work assistant—all rapidly iterating and receiving broad customer acclaim. Amazon Q, in particular, is growing rapidly, with major enterprises like Mondelez, EchoStar, 3M, Southwest Airlines, and DXC adopting it as a work-focused, data-secure AI companion.

The chip business continued triple-digit growth, with revenue commitments for Graviton and Trainium up more than 20x year-over-year. AI provides incremental growth for Graviton since post-training reinforcement and agent tool use primarily run on CPU rather than AI accelerators—Graviton is the most powerful CPU chip, offering 30–40% better price-performance than alternatives. In Q3, customers saved $2 billion by using our custom chips. We've open-sourced TorchNeuron so developers can run existing PyTorch code on Trainium with zero rewriting—most developers use PyTorch already.

New-generation chips haven't diminished prior-gen demand: After Graviton 5 launched, prior-gen Graviton instance backlogs doubled quarter-on-quarter; prior-gen Trainium spot instance prices saw premiums over H1 levels. With such intense Trainium demand, customers keep urging us to offer it externally. Our plan is to begin outside sales by year-end 2027.

Overall, AWS capacity demand remains exceptionally strong. Customers are still in the early days of large-scale inference apps, but inference now constitutes over half—and continues rising—of AWS AI revenue; meanwhile, training revenue also remains sizable. We maintain our 2026 capex forecast of ~$220 billion, with most flowing to AWS, and expect a significant further rise in 2027 spending—all with a visible path to attractive financial returns, as I highlighted on our last call.

AWS's current annualized revenue run rate is $184 billion, but the opportunities ahead are enormous. With our innovation pace, fast-evolving core and AI features, strong customer intent to migrate workloads, and ever-growing capacity, we believe AWS will achieve 40%+ growth in both 2027 and 2028—becoming a trillion-dollar revenue business in the next 5–10 years. We are at the very beginning of this inflection point, our strategy is working, and I believe this momentum will persist as far as we can foresee.

Retail Business

I think some have yet to fully appreciate the progress and scale this business has achieved in recent years. I’m unaware of any retailer that reaches Amazon's hundreds-of-billions scale, maintains double-digit operating margins (including advertising), and is still growing this fast.

This year-to-date, sales units are up nearly 15% year-over-year, on a massive base—over 30 billion items sold last year. In absolute terms, we’ve added over $90 billion in goods sold (GMV) in the last twelve months—much faster than competitors, adding twice as much GMV each quarter as the #2 U.S. competitor, taking share across all retail verticals.

In recent years, both North America and International businesses raised operating margins nearly ten points, enabling continued reinvestment in customer experience. But this is still early—about 80% of retail sales still happen in physical stores; that will shift to ecommerce over time. We're building fast, efficient fulfillment networks to provide better shopping and create ample free cash flow and ROI, thus moving this transition forward.

Our strategic focuses—faster delivery, lower unit costs, fresh and pharmacy, and AI shopping—reinforce each other, and are all advancing aggressively. We continue heavy investment in same-day fulfillment, Amazon Now, and Amazon Prime Air, driving record speeds. Q3 was our fastest ever globally, with over 1.5 billion items delivered same-day or next-day to Prime members just in Q3.

Amazon Now is scaling quickly, now covering over 450 towns and cities worldwide, with items ordered via Now up 10x year-over-year. For the fourth year in a row, we expect to reduce per-unit fulfillment costs, allowing us to offer competitively priced, high-assortment fresh food and essentials.

Speed improvements fuel rapid grocery expansion (including fresh and essentials), with goods sold reaching nearly $250 billion over the past year. In Q3, fresh and essentials grew about three times faster than other categories, now making up about half of units sold. Customers who order fresh food same-day buy nearly 7x as many essentials as non-fresh shoppers; fresh grocery customers also buy home, apparel, and other non-groceries about four times as often, spending twice as much as non-grocery customers on these other categories.

Amazon Pharmacy is also seeing rapid growth: Customer count is up over 60% year-over-year, with growth rates accelerating throughout 2026; repeat purchasers are up over 50%. JD Power has ranked Amazon Pharmacy as the top mail-order pharmacy for three straight years. Our capabilities keep improving, now with 3-hour prescription delivery in over 350 cities.

On AI shopping, over 400 million customers have used Alexa Shopping over the past year, with Alexa interactions up 570% year-over-year; U.S. Alexa Shopping customers spend 50% more per order than non-Alexa customers. Alexa assists in product research, comparisons, price history, personalized gift ideas, tracking deals, and pricing. While there may be cross-platform agents in future, we believe Alexa for Shopping will remain the top choice for Amazon shoppers—ever-present to help you browse, get inspired, answer product details, and place orders via your purchase history, address, and payment info—all by voice. It's a delightfully convenient experience.

Advertising Business

Ad revenue in Q3 was $19.6 billion, up 12% year-over-year. Growth was slowed by several timing issues this quarter:

First, most countries moved Prime Day to Q2, shifting substantial sales and ad revenue from Q3 to Q2, providing a tailwind for Q2 and a headwind for Q3—about 850 basis points of impact on ad revenue growth.

Second, Q2 2026 included the debut NBA on Prime season, while prior Q2 had none, and Q3 had no NBA (which begins in Q4).

Third, in Q3 2025, we adjusted search engine marketing projects to drive more traffic to Amazon's search pages—providing better customer experience and monetization, accelerating ad growth last year, but now presenting a tough comparison base this Q3.

We expect Q4 ad revenue growth to return to high teens percentages year-over-year.

As for whether external shopping agents impact Amazon's ad business—the answer is no. Few consumers actually shop that way; experiences are not yet mature and much less convenient than researching and buying directly on retailers’ sites.

In Q3, our ad team launched many new products. Most notably, Amazon Ads Agent integrated our entire ad product ecosystem on an AI-centric platform, leveraging trillion-scale shopping, browsing, and streaming signals to support advertisers through insights, answers, optimization, and even execution. We also launched full-funnel campaigns, aggregating sponsored ads, display, video, and streaming TV in a single campaign. Advertisers just set an initial budget and run length; Amazon AI plans, executes, and optimizes the rest.

Other Business Highlights

Prime Video has become a major business, with annualized revenue run rate in the tens of billions, now profitable on its own. We launched hits such as "Carrie," "The Summer I Turned Pretty," "The Boys," "Gen V," "Doctor Who," and "Fallout," plus breakout theatrical movies like "The Burial" and world-class sports rights, including Thursday Night Football (2026’s debut hit records), NBA on Prime, NASCAR, celebrity events, and the UEFA Champions League.

We've just launched the strongest hardware portfolio ever—across Echo, Kindle, Alexa tablets, Fire TV, Ring, and Blink, with 34 new devices. It's early in the launch cycle but momentum is strong, and we look forward to great holiday season experiences.

Finally, Amazon Leo commercial service is still on track to launch initial services this year. With re:Invent and the holidays coming, we’re entering one of the most exciting phases of the year. Thanks to all Amazonians for your hard work and relentless innovation, making our customers' lives better and more convenient every day.

Amazon is building an unprecedented business map. Never before has a company reached $1 trillion in annual revenue, but Amazon is now on the path to two separate trillion-dollar businesses—AWS and retail. It's extremely rare, and we look toward the future optimistically.

Now Brian will speak.

Senior Vice President & Chief Financial Officer Brian Olsavsky Remarks

Revenue Overview

Worldwide revenue was $201.9 billion, up 12% year-over-year, or 13% excluding FX impacts. Global operating profit was $28.1 billion, a record-high 13.9% operating margin.

North America and International Segments

First, a reminder: Prime Day events were rescheduled this year, with most major markets (including the US) moving to Q2, and 2025 Prime Day planned for Q3. Excluding Prime Day timing, Q3 2026 growth would be about 400 basis points higher. Q3 saw strong customer engagement, continuing H1's momentum.

North America: Q3 revenue was $112.9 billion, up 6% year-over-year.
International: Q3 revenue was $43.1 billion, up 8% year-over-year excluding FX.
Global paid units sold: up 12% year-over-year.
Prime remains a critical pillar, with hundreds of millions of members responding positively to our vast selection, low prices, and record delivery speeds. Over the past year, Prime members' same-day fresh grocery orders tripled monthly.

Profitability

North America: Operating profit $8.7 billion, 7.7% margin.
International: Operating profit $1.6 billion, 3.7% margin.
In fulfillment, the team overcame transport cost pressure from fuel and linehaul inflation. Despite these headwinds, Q3 saw overall per-unit fulfillment cost reduced through optimized inventory, shorter delivery distances, and package consolidation. Fulfillment costs rose 11%, while paid units rose 12%; excluding fuel/linehaul inflation, cost growth was 8%, about 400 basis points less than units.

Robotics and AI have delivered quantifiable gains in efficiency and worker safety, with improvements expected to compound over time. In 2026, we doubled AI arm deployments (Cardinal, Sparrow) and are developing robot foundation models to accelerate programming and teach robots more complex tasks. The goal: large-scale practical deployment. Looking ahead, we see major opportunities for network-wide productivity improvements, long-term per-unit cost reductions, and ever-faster delivery. Although operating margins may fluctuate quarterly, we strive for ongoing unit cost improvement.

AWS Segment

AWS Q3 revenue was $46 billion, up 39.3% year-over-year, with both core and AI services contributing. Sequential revenue up $3.8 billion; annualized run rate at $184 billion. Customers are steadily migrating and expanding AWS usage; those seeking to leverage AI are accelerating cloud migration. We’ve observed a strong positive linkage: As AI spend increases, so does core AWS service consumption. We expect this linkage to strengthen as more AI workloads move to production, boosting compute, storage, and networking demand.

AWS operating profit was $17.8 billion, with a 38.7% margin, reflecting strong performance and continued efficiency. Custom chip, network, software, and process improvements optimize capacity and boost infrastructure cost-effectiveness. As always, margins may fluctuate with different investment cycles. AI products are being adopted rapidly, with AI revenue up triple digits year-over-year.

Capital Expenditure

Q3 capex was $75.9 billion, mainly for AWS and AI infrastructure to meet strong customer needs. We are convinced AI is a huge opportunity, set to drive long-term revenue, free cash flow, and exceptional returns.

Q4 Financial Guidance

Net sales: Expect $240–$247 billion.
FX impact: About a 60 bps headwind on revenue growth based on current rates.
Operating profit: Expect $28.5–$33.5 billion.
Note on Q4 guidance: Amazon Leo commercial service is proceeding to initial launch this year. Starting Q1 next year, some related costs will be capitalized (production/launch). Since this is later than forecast on our prior call, most Leo costs will continue to expense in Q4.

We’re fully prepared for the holiday season and are supporting millions of AWS customers worldwide. Thanks to the entire Amazon team for your hard work and customer-centricity. We believe improving the customer experience is the only reliable path to lasting shareholder value.

We will now begin the Q&A session.

Q&A Session

Question 1: Justin Post (Bank of America)
Q: AWS acceleration is exciting, but sequential growth slowed; what factors are behind this? Also, Brian, how do fuel costs impact Q4 guidance?

Andy Jassy (AWS Growth):

We’re very satisfied with 39.3% growth—that's impressive for a business at a $184 billion annual run rate, the best in nearly five years and the sixth straight accelerating quarter. The growth drivers are multiple.

On AI, we have a full stack—from Sagemaker AI for custom model-building, to Bedrock for secure, cost-effective top models, to Strands and Agent Core for building your own agents, to turnkey agents like Kiro (programming), Connect (call centers), Transform (migration), and Q (AI assistant)—all seeing rapid customer adoption. On chips, we’re ahead: For major enterprise AI at production scale, especially as inference grows, customers want best price-performance, and Trainium beats alternatives. As AI proliferates, demand for reinforcement, post-training, and tool use grows, with Graviton—our top CPU—offering 30–40% better price-performance than x86. All those support significant growth.

Quarterly capacity adds vary, but forecasting 40%+ growth in 2027/28 from a $184 billion baseline is huge. We believe AWS can be a trillion-dollar business in 5–10 years—an exceptionally rare opportunity.

Brian Olsavsky (Fuel Costs):

I don't have a precise number on fuel/linehaul inflation, but Q4 will be slightly higher than Q3. At the beginning of the year, we added a fuel surcharge for FBA sellers, covering part of the cost; the rest hits retail inbound/outbound shipping. Q4 will be a bit above Q3 for these costs.

Question 2: Brian Nowak (Morgan Stanley)
Q: What supports your conviction in AWS hitting $1 trillion revenue? Also, what are the 2027 robot tech milestones?

Andy Jassy (AWS Path to $1 Trillion):

Growth rate, base revenue, and forward forecasts for 2027/28 all support it. We have good visibility on large enterprise deals (hundreds of MWs to GW reservation), which must be booked well in advance, so we see demand for the next 8–10 quarters pretty clearly. Extrapolating the growth rapidly increases the numbers.

I also don't see usage fading soon—enterprise customers are just beginning to deploy inference at production scale. One variable is inference cost; for usage at our envisioned scale, we must keep lowering costs, a key reason for Trainium's development. Over the past two years, inference costs have fallen a lot—and should continue declining.

Security will be a huge use case going forward—current models are surfacing vulnerabilities people never found; some estimate a big chunk of token consumption over the next year or two will be just for security. This is atop code-writing and other established scenarios. Add in agent/assistant growth—consumer or enterprise (like Amazon Q)—and I believe nearly every enterprise production app will eventually re-emerge as an agent-enabled version—for speed/agility and security. The AI future holds so many possibilities. While we can't pinpoint the next 5–10 years exactly, AWS as a trillion-dollar business is our real trajectory.

Andy Jassy (Robotics Milestones):

We already have over 1 million robots in the fulfillment network, which significantly improves costs, productivity, and safety—our fulfillment safety measures are at all-time highs, with major gains in the past five years, thanks largely to robotics. Still, we are early in both deployment and tech development.

I believe more repetitive tasks in fulfillment centers will be performed by robots; we’re also building variants that may handle the "last 100 yards" to homes. Our bar is whether a robot can drive equivalent or better productivity, safety, cost, and speed; if it makes economic sense, we’ll scale it. New fulfillment center designs, like same-day centers, are robot-heavy. Robotics to date have helped a lot, but the next five years will be transformative by comparison.

Question 3: Mark Mahaney (Evercore ISI)
Q: Will AWS's margin structure stay stable for the next two years? How is grocery coverage driving Prime member spend?

Brian Olsavsky (AWS Margin Outlook):

We don't give explicit AWS margin guidance but some context: Q3's high 38.7% margin reflects years of operational discipline, efficiency, capacity optimization, cost control, and custom chip adoption—not happenstance. But there are volatility factors: AI’s explosive growth depresses margins for now (AI lags the core in profitability), and component pricing is a headwind. AI margins are improving (as Andy said, ahead of where core was at a similar stage), but are still lower than the core. Overall, margin will always be multivariable; we'll keep working to improve costs and offset dilution from new products.

Andy Jassy (Grocery Business):

We're as excited as you are about grocery. We've long run a big grocery business but focused on the center-of-store shelf—consumables, canned food, pharmacy, beauty. We have Whole Foods, a leading organic grocer with top-tier growth and profit, rolling out new formats and Daily Shop locations with great feedback.

We've always known that to serve customers as we want, mainstream “mass” grocery is essential—especially fresh. When you compare the categories we offer versus where customers actually spend their total household dollar, the biggest gap is grocery, and mass grocery starts with fresh.

Therefore, we're excited about our same-day fresh fulfillment progress. Essentials are growing three times faster than other categories, now about half our sales units. Same-day fresh customers buy 7x as many essentials as non-fresh; this helps close the gap between "total customer spend" and "Amazon spend." Meanwhile, grocery shoppers buy 4x more frequently in non-grocery categories (home, apparel, etc.) and spend over double compared to non-grocery shoppers. These are significant effects, and continued progress in fresh/grocery will help make us the default in more shopping occasions.

Question 4: Eric Sheridan (Goldman Sachs)
Q: What’s Bedrock's market opportunity? How are you balancing internal/external Trainium sales?

Andy Jassy (Bedrock Market Opportunity):

Twenty years serving tech customers have taught us: No "one tool rules them all." In databases and analytics, early leaders were never the only choices. For AI, the same applies—people first bet on TensorFlow, now PyTorch is more mainstream. Models won't be different.

I believe six to eight leading models will coexist in the next 2–3 years, each with strengths—on inference, code, vision, latency, cost, etc.—all meeting 95%+ of workload needs. Most customers will use several models at once, often mixing them inside a single generative AI app.

Customers want broad choice of models, price-performance, strong security and access controls, and the flexibility to switch as models progress. That’s what Bedrock offers. We have both closed-source models (Anthropic, OpenAI) and open (Kimi, GLM), and strong demand for more Western open models. This market is enormous—potentially as large as EC2 as a cornerstone AWS service.

Andy Jassy (Trainium Internal/External Balance):

We continue to evaluate this. Only two companies in the world operate both a significant cloud business and manufacture external-grade chips at this scale—balancing this is unique.

All advanced chips are fabricated at the same foundry, which has finite capacity; each company must determine how much to allocate to their own use versus external sales.

The tradeoff: Selling externally recognizes revenue up front as chips ship; using chips in our cloud brings revenue and cash flow over 5–6 years, ultimately yielding more. So it's a careful balance.

We approach this differently from the other major firm. We plan to start selling Trainium externally by year-end 2027, and may revisit the balance every year. Both channels are important. Our ability to build high-performance, scalable custom chips is a strategic advantage—optimizing our own economics and establishing an independent external chip business. The chips and cloud infrastructure are fundamentally different models.

Question 5: Doug Anmuth (J.P. Morgan)
Q: What size capex is required to sustain 40%+ AWS growth for 2+ years? How are payback periods for AWS servers, networks, data centers trending?

Andy Jassy:

On payback period—we’re indeed shortening them. Different asset classes (AI vs. classic compute, different financing) vary, but overall we’re extending useful lives and shortening paybacks, with consistent progress.

As for the capital expense required to support those growth rates, it's: substantial. But we think this is a historic, once-in-a-lifetime opportunity. With a $184 billion annual run rate and 39% growth, then two more years of 40% growth, ending up with a $1 trillion business—to maintain leadership, we will invest. If this opportunity could be spread over 25 years, capex would be gentler, but that's not reality. We have the capacity, and this will permanently change Amazon’s scale—not just revenue, but free cash flow, keeping us ahead.

Question 6: Ross Sandler (Barclays)
Q: As the Bedrock token market (inference as a service) scales, with most workloads running on Trainium, how will this business transition affect AWS revenue growth and especially margins?

Andy Jassy:

You're right, most Bedrock workloads do run on Trainium—because we're very pleased with Trainium's price-performance: It lets us keep customer pricing low while protecting our desired margins.

Broadly, several cost structure elements matter in the cloud industry.

First, data center ownership: Owning or long-term leasing is about 3,500 basis points better for operating margin than colocation or neocloud models.

Second, hardware: Using the same hardware as competitors means average cost structure; using differentiated hardware confers cost advantage.

Most of our capacity is owned or leased, which beats competitors running colo or neocloud models. With proprietary chips, we have structural cost advantages that benefit both customers and our own economics. These will continue to help AWS’s overall profit profile.

Thanks for joining today’s call. The replay will be available on our Investor Relations website for at least three months. Thank you for your interest in Amazon and we look forward to connecting next quarter.