AWS Accelerates to 39% and Guides Above 40% Through 2028; the Market Is Pricing the 2027 Capital Bill Instead

Results vs. Consensus

Amazon delivered the strongest operating quarter in our coverage and missed the two headline consensus numbers anyway. Revenue landed $56M below the top of the company's own range and 0.2% below the Street, which is a rounding difference in a quarter that lost Prime Day to Q2. EPS missed by 7 cents, and the whole miss is non-operating: consensus evidently assumed positive other income, and Amazon reported a $703M loss there plus a tripled interest bill. The figures that decided the stock's direction were not the headline consensus lines: a ~42–43% buy-side bar for AWS, advertising growth of 12% against roughly 18%, and gross capex about $20B above estimates.

Q3 2026 Scorecard

Year-over-Year Comparison

Sequential Comparison

Quality of Beat

• Revenue: Timing-distorted, and better than it reads. Reported growth of 12.1% (13% ex-FX) excludes Prime Day, which moved into Q2 for most large countries after sitting entirely in Q3 in 2025. Management put the effect at approximately 400 bps, which puts underlying growth near 16% as reported and 17% ex-FX. That is in line with or slightly above Q2's underlying rate. Paid units grew 12% against 17% in Q2, the same Prime Day distortion, and year-to-date units are up nearly 15%. The revenue composition shifted hard toward services: net service sales grew 19.4% while product sales grew 1.7%.
• Margins: Broad-based on the clean comparison. Net sales less cost of sales rose 238 bps to 53.2%, driven by mix: AWS is 22.8% of revenue against 18.3% a year ago, and service sales outgrew product sales by more than 17 points. Below that line, fulfillment expense grew 8.4% on 12% unit growth. AWS supplied 63.3% of consolidated operating income. North America margin of 7.7% is up 87 bps on the ex-FTC basis, and International rose 73 bps to 3.7%. Technology and infrastructure expense grew 20.2%, faster than revenue, and that is where the depreciation is landing: D&A rose 24.7% to $20,949M, or 10.4% of revenue against 9.3% a year ago.
• EPS: The miss is below the operating line. Pre-tax income excluding other income was $27,572M. Taxed at our normalized 20%, that is about $2.02 per share, essentially flat on Q2's $2.01 by the same method and up roughly 23% on the year-ago $1.64 after the $4.3B of charges are added back. At the reported 23.0% tax rate it is $1.94. Consensus at $1.96 evidently assumed positive non-operating income. Interest expense of $1,691M is now running at an annualized $6.8B, and it will keep rising with the debt.
• Free cash flow: The weakest line again, and the quarter's own number is worse than the trailing one. Q3 operating cash flow of $42,422M against net cash capex of $75,878M is a single-quarter outflow of ($33,456M). Trailing operating cash flow grew 28.8%, while trailing capex grew 81.8%. Equipment acquired but not yet paid rose only $853M in the quarter against $10,700M in Q2, so, unlike Q2, almost none of the quarter's additions were left unpaid. On the trailing accrual basis, deducting the $26,779M twelve-month increase, free cash flow is near ($69.1B).

Segment Performance

Segment Mix: Q3 2026

Revenue by Line of Business: Q3 2026

AWS

Revenue of $45,984M grew 39.3%, the sixth consecutive quarter of acceleration and the fastest rate in nearly five years, on an annualized run rate of $184B. Segment operating income of $17,773M rose 55.4% at a 38.7% margin. AWS is now 22.8% of revenue and 63.3% of operating income. The sequential dollar add of $3,752M was the second-largest in AWS history, behind only Q2's $4,645M, but it was about $900M smaller than Q2's. Management's explanation is that capacity lands in uneven blocks.

"Despite lapping a comparable year-over-year period where we added a lot of capacity, AWS grew 39.3% year-over-year, accelerating for the sixth straight quarter, our fastest growth in nearly five years. Backlog is $578 billion, growing triple digits year-over-year and well-diversified. We're installing an AI accelerator about every 14 seconds, and almost as fast as we install capacity, customers are gobbling it up." — Andy Jassy, CEO

The margin read is cleaner than the headline suggests. Reported margin fell 71 bps from Q2's 39.4%, but Q2 included most of a ~$600M energy-derivative gain, and on the roughly 38.0% clean Q2 figure AWS margin rose about 65 bps. That happened while depreciation is rising, AI mix is growing, and memory costs are inflated, all of which management cited as headwinds. Core compute and storage grew at "their fastest rates in several years," which supports the flywheel argument from Q2: AI demand pulls core consumption with it.

Assessment: Fundamentally the best AWS quarter of the cycle, and the stock fell on it because the bar had moved to 42–43%. A 39% grower adding $3.8B a quarter at a 38.7% margin, with a backlog of 3.6 times trailing revenue, does not have a demand problem. The slower dollar add is a capacity-timing question, and management said 8 to 10 quarters of capacity are planned. The thing to watch is whether Q4 restores a $4B-plus add. If it does not, the uneven-capacity explanation weakens.

North America

Revenue of $112,852M grew 6.2%, depressed by Prime Day moving into Q2. Operating income of $8,723M rose 82% as reported and 19.7% against the $7,289M ex-FTC base, for a 7.7% margin. That is 87 bps above the 6.9% ex-FTC year-ago margin and 39 bps above Q2's 7.3% clean margin. One caveat limits the year-over-year read. Part of the year-ago $1.8B severance charge also sat in this segment, and Amazon never disclosed the split. If all of it had been North America, the clean comp would be 8.6% and this quarter would be down 82 bps. The true comparison lies between those bounds. The sequential improvement is unambiguous, and it came in a quarter without Prime Day's volume.

"Shipping costs increased 11% year-over-year versus a 12% increase in units. Excluding fuel and line haul inflation, shipping costs grew 8%, or approximately 400 basis points slower than unit growth." — Brian Olsavsky, CFO

The cost evidence is the better news. In Q2, management said only that shipping costs ex-fuel grew more slowly than units. This quarter it quantified the gap at about 400 bps, and fulfillment expense grew 8.4% against 12% units. Fuel and line-haul are still a headwind: the CFO said the Q4 impact is "a little bit higher than Q3," and the FBA surcharge introduced earlier in the year recovers only part of it.

Assessment: The downgrade trigger set after Q2 was a second consecutive clean-basis margin decline. It did not happen: the margin rose sequentially on a clean basis and rose year over year on the ex-FTC basis. We keep the pillar at risk rather than restoring it. The severance split leaves the year-over-year comparison uncertain. The release also announced a $1.5B-plus investment in U.S. operations pay ($20 an hour minimum starting pay for full-time staff), with no timing and no margin quantification.

International

Revenue of $43,108M grew 5.4% as reported and 8% ex-FX. Operating income of $1,580M was up 31.8% at a 3.7% margin, against 2.9% a year ago and 4.1% in Q2. Currency took $125M off segment operating income, the largest FX drag in the six-quarter table, and adding it back puts the margin at about 4.0%. Growth of 8% ex-FX this quarter suggests that much of Q2's 15% was Prime Day timing.

Assessment: Steady and immaterial to the thesis. The segment is 21% of revenue and 5.6% of operating income, with margin up 73 bps year over year. The trailing margin of 3.3% is moving in the right direction, but slowly, and a single quarter of FX noise does not change that.

Advertising

Advertising revenue of $19,637M grew 10.9% as reported and 12% ex-FX. That is down from 26% in Q2 and 22% a year ago, and it is the first time in seven quarters that the line fell short of consensus. Management gave three timing causes and a size for the largest. Prime Day's move into Q2 accounted for 850 bps of the sequential change in the year-over-year growth rate. NBA on Prime ran in Q2 2026 with no prior-year comparison and does not resume until Q4. And a search-engine-marketing change made in Q3 2025, which drove more traffic to Amazon search pages, is now being lapped.

"The natural question is whether agentic shopping from external horizontal agents is meaningfully impacting Amazon Ads. The answer is no. Few customers are doing much shopping this way yet, as the experience is still not great nor more convenient than coming to a retailer directly to research or one-click order there." — Andy Jassy, CEO

Amazon also put the recovery in the release itself: year-over-year advertising growth "in the high teens" in Q4. On the Q4 2025 base of $21,317M, that implies roughly $24.9B to $25.4B. The product cadence supports it: Amazon Ads Agent, full-funnel campaigns across sponsored, display, video and streaming TV, and NBA returning in Q4. Trailing advertising revenue is $78.0B.

Assessment: The explanation is specific and checkable, which is what separates a timing miss from a deflection. Adding back the 850 bps puts growth near 20%, and the NBA and SEM effects explain most of the remaining gap to Q2's 26%. Management also answered the agentic-disruption question unprompted and directly. The pillar stays on track, but the Q4 high-teens guide is now a hard test. Growth below 15% in Q4 would move it to at risk.

Key KPIs

Key Topics & Management Commentary

Overall Management Tone: The most forward-committal call in our coverage. Management gave a two-year AWS growth floor, a dated Trainium external launch, and a trillion-dollar framing for two separate businesses, and it was prepared and on-message throughout. The commitment ran in one direction, though. Every new forward number was a revenue or demand number, and the cost side got only the standard Q4 guide and an unchanged 2026 capex plan. Every cost question (2027 capex, AWS margin over two years, the capital required for 40% growth) got a framework or a refusal, and the only funding statement was a one-sentence assurance that the plan can be funded.

1. AWS: The Sixth Acceleration, and a Smaller Dollar Add

The quarter's tension is in one pair of numbers: growth accelerated to 39.3% while the sequential dollar add fell to $3.8B from $4.6B. Both can be true because the rates are measured on different bases. Sequential growth was 8.9% this quarter against 6.9% in Q3 2025, so the year-over-year rate rose even though the dollar add fell from Q2's record. Management also noted that the comparison was not an easy one: it was "lapping a comparable year-over-year period where we added a lot of capacity." The opening question of the call went straight to the slower sequential add, and the answer was that capacity arrives unevenly (see Analyst Q&A below).

"As we launch new generations of chips, demand for prior generations remains strong. After our launch of Graviton5, backlog for prior generation Graviton instances still more than doubled quarter-over-quarter. Short-term capacity for prior generation Trainium instances sell at a premium compared to the first half of the year." — Andy Jassy, CEO

The demand evidence is not in doubt: a backlog up $82B in one quarter, capacity that customers take "almost as fast as we install" it, and prior-generation instances that are still in demand after a new generation launches. In a supply-constrained business, the sequential add measures how much capacity came online, not how much customers wanted.

Assessment: This is an expectations miss, not a fundamental one, and the explanation fits the facts. The risk is that a capacity-gated growth rate is only as reliable as the construction schedule. Amazon has declined for two quarters to give the first-half versus second-half capacity split that would let the Street model the adds, so the growth rate cannot be checked against a supply plan from outside.

2. The First Multi-Year AWS Outlook: Above 40% in 2027 and 2028

Amazon does not guide segments, and in our coverage it has not given a forward AWS growth rate before. This quarter it gave two, in the press release and on the call: growth above 40% in each of 2027 and 2028. Management tied the visibility to how large AI capacity is sold. Blocks of hundreds of megawatts to a gigawatt are reserved well in advance, so the 2027 and 2028 books are largely known.

"With our continued pace of innovation, new core and AI functionality, enterprises wanting to run their core and AI workloads in AWS, and the amount of capacity that we're adding each year, we believe that AWS can grow north of 40% in each of 2027 and in 2028 and be a trillion-dollar revenue business in the next five to 10 years." — Andy Jassy, CEO

The arithmetic is large. From an FY26 AWS base of roughly $176B (three reported quarters plus a Q4 estimate), 40% a year gives about $246B in 2027 and $345B in 2028. The trillion-dollar figure itself was raised as an aspiration in Q2, so it is the time frame that is new. Management also extended the framing to retail, saying Amazon may have "two" trillion-dollar revenue businesses in 5 to 10 years.

Assessment: The 2027–2028 outlook is the most valuable disclosure of the call, because it is falsifiable and near-term. A management team that has beaten the top of its operating income guide by $3.5B and $1.6B in consecutive quarters does not put out a two-year floor lightly. We model 40% for FY27. The trillion-dollar statement is direction, not forecast, and we give it no weight in valuation.

3. The Backlog Adds $82B in a Quarter, to $578B

Backlog rose to $578B from $496B in Q2 and roughly $200B a year ago, a 16.5% sequential increase. It now equals 3.6 times trailing AWS revenue and 3.1 times the annualized run rate. Management called it "well-diversified," which answers a concentration question raised after Q2, when the breadth language from Q1 was not repeated. The release named new agreements with Meta, NVIDIA, OpenAI, Salesforce, Oracle and others.

The quarter-on-quarter dollar increase in backlog ($82B) was about 22 times the quarter-on-quarter increase in revenue ($3.8B). The book is filling much faster than capacity can convert it, so the gap between contracted demand and delivered revenue is still widening.

Assessment: The backlog is still the strongest single argument for the capex program, and "well-diversified" is useful even without numbers. Two things limit it. The term structure of the $578B is undisclosed, and several of the named new counterparties are themselves AI labs or AI infrastructure companies with their own funding needs. Diversified across AI-heavy counterparties is not the same as diversified across the enterprise base.

4. Trainium Goes External in Late 2027, and the Run-Rate Disclosures Stop

Selling Trainium outside AWS moved from "a real chance we'll do that in the future" in Q2 to a dated plan: late 2027. Management described the business as "very significant" and a structural advantage. It also explained the trade-off it manages each year inside a fixed foundry allocation: external sales recognize revenue at shipment, while internal deployment earns revenue and free cash flow over five to six years.

"Finally, our chips business continues to grow triple-digit percentages year-over-year, and revenue commitments for Graviton and Trainium have each increased more than 20x year-over-year. ... With Trainium demand so strong, customers continue encouraging us to make it available outside of AWS. We intend to do so late in 2027." — Andy Jassy, CEO

The disclosure trend runs the other way. Q1 sized the chips business at a $20B-plus run rate and a $50B standalone equivalent. Q2 gave a run rate above $25B, without the standalone figure. Q3 gave neither, only "triple-digit" growth, revenue-commitment multiples, and more than $2B of customer savings in the quarter. The AI revenue run rate (above $25B in Q2) was also dropped, replaced by the statement that inference is "well over half" of AI revenue.

Assessment: The external-sale date confirms the silicon pillar and partly meets the bull-case trigger we set at Q2, a disclosed timeline. The revenue starts in late 2027, so it is an FY28 model input. Dropping two run-rate figures in one quarter is a negative signal on disclosure, not on the business. Commitments up more than 20 times year over year are consistent with very strong growth, but they are not a revenue number.

5. Bedrock on Trainium: Where Management Says the Margin Comes From

Management made its most explicit structural margin argument yet, in response to a question on whether a shift to inference-as-a-service changes AWS economics. Most Bedrock workloads run on Trainium. Most AWS capacity is owned or long-leased rather than colocated. Management quantified the difference against neocloud capacity.

"Neoclouds are about 3,500 basis points worse in operating margin than if you long lease them. Then you look at what is the hardware on top of which they are running, and if you are using the hardware that everybody else is using, you will have a comparable cost structure to what everybody is doing. If you have the ability to use hardware that is different, you will have an advantage." — Andy Jassy, CEO

Bedrock is described as the fastest-growing multi-billion-dollar service in AWS history, with Fortune 500 spend up nearly 7 times and token volume up nearly 17 times year over year. Management also said it "could be a service as large as EC2." The model roster now includes OpenAI's GPT-6 family alongside Anthropic's models, plus open-weight models from Moonshot AI and Z.ai.

Assessment: This is the mechanism behind a 38.7% margin during an AI buildout, and it is coherent: own the building, own the chip, and sell tokens rather than bare metal. The 3,500 bps figure is management's own estimate and cannot be verified from outside. What can be verified is the segment margin, and it has held above 37% for three consecutive quarters while AI mix rose.

6. AWS Margin at 38.7%: The Mix Warning Arrives

Asked whether high-30s AWS margins are stable over two years, the CFO declined to guide. He then gave the clearest statement yet of the downward forces: AI margins are lower than core, AI is growing faster, and component pricing is a headwind. In Q2 the framing was "margins will fluctuate". This quarter it was a named mix issue.

"In fact, the outsized growth of AI puts margin pressure on the business as a whole because the margins are lower there. We are happy with the progression in both core and AI margins, and as Andy said earlier, they are ahead of where the core was at a similar point in time, but they are still lower, so there is going to be that mix issue. There are also headwinds from component pricing that we have talked about." — Brian Olsavsky, CFO

The CEO's prepared remarks gave the other side: AI margins are "rapidly improving," "well ahead" of where core was at a similar stage, "even with heightened component costs like memory and significant depreciation from the CapEx we're investing."

Assessment: Read together, the two statements say AWS margin peaks somewhere near here and drifts lower as AI becomes the majority of growth. We now model about 36% for FY27, and the call makes that more likely. In Q2, mix was one item in a list of reasons margins fluctuate. This quarter it was named as a directional pressure, which moves Bear 5 from a P&L question management would not discuss toward one it has started to describe.

7. Capex: $220B Held, 2027 "Meaningfully More," and a Q4 Number That Does Not Fit

Full-year 2026 cash capex stayed at approximately $220B. For 2027, management said only that it will be "meaningfully more," with "a clear line of sight to attractive financial returns." Asked directly what capex 40% growth requires, the CEO called it "meaningful" and a "once in a lifetime opportunity," and gave no number.

"We still expect 2026 cash CapEx of approximately $220 billion, the majority for AWS, and expect to spend meaningfully more in 2027 with a clear line of sight to attractive financial returns, as I laid out on our last call." — Andy Jassy, CEO

The Q4 arithmetic does not reconcile. Net cash capex was $43,234M in Q1, $53,076M in Q2 and $75,878M in Q3, so nine months total $172,188M. The ~$220B plan leaves about $47.8B for Q4, a 37% sequential drop in the quarter before a year of "meaningfully more." There are explanations. In Q3, almost all equipment additions were paid in cash (equipment acquired but not yet paid rose only $853M, against $10,700M in Q2), so cash capex tracked additions more closely than in Q2, and "approximately" covers a few billion. But management did not reconcile it.

Assessment: The 2026 number is effectively a cash-timing figure now, and the real question is 2027, which the market will price on the Q4 call. We model FY27 net cash capex near $270B, about 17% to 23% above 2026. The implied Q4 step-down is more likely a cash-timing effect than a real slowdown, so we expect FY26 to land above $220B rather than Q4 to drop to $48B.

8. Free Cash Flow, Debt, and the Interest Line

Trailing free cash flow fell to ($42,357M), from ($7,604M) in Q2 and +$14,788M a year ago. Q3 alone was ($33,456M). Long-term debt rose $44.2B in the quarter to $173,125M, with $48.1B of long-term borrowing in Q3 and $115.1B year to date. Cash and marketable securities of $114.2B leave net debt of about $58.9B. Interest expense was $1,691M, up 214% year over year and 29% sequentially. The investing section also shows another $23,986M under acquisitions and non-marketable investments, bringing the year-to-date total to $63,753M.

"Turning to our cash CapEx, which was $75.9 billion in Q3. This primarily relates to AWS and AI as we invest to support strong customer demand. We continue to make significant investments, especially in AI, as we believe it to be a massive opportunity with the potential to drive long-term revenue and free cash flow, as well as return on invested capital." — Brian Olsavsky, CFO

Assessment: Bear 1 is fully materialized and still escalating, and 2027 will be worse on capex before it is better on revenue. The leverage is serviceable, at $6.8B of annualized interest against $168.3B of trailing operating cash flow. The issue is visibility, not solvency: the investing outflow outside capex was about a third of capex in Q3 and 37% of it year to date, and it has gone unexplained for two consecutive quarters.

9. North America Margin: The Downgrade Trigger Did Not Fire

After Q2 we said a second consecutive clean-basis decline in North America margin would move the rating to Hold. The test result is 7.7%: up 39 bps on Q2's 7.3% clean margin, and up 87 bps on the year-ago 6.9% ex-FTC margin. Trailing North America margin reached 8.2%, the highest in the six-quarter table. Management said Amazon is on track to lower cost to serve per unit for a fourth consecutive year.

"We're on track to reduce cost to serve per unit for the fourth consecutive year, which enables us to continue meeting or beating competitors' prices, and cost effectively deliver a broad selection of fresh groceries and Everyday Essentials." — Andy Jassy, CEO

The operating levers broadly delivered. The robotic-arm fleet doubled in 2026, which meets Q2's commitment that it would more than double only at the line. More than 1 million robots are now in the network, and new same-day facility designs are "very heavy on robotics." Shipping costs ex-fuel grew about 400 bps slower than units.

Assessment: A pass, but a qualified one. The sequential improvement without Prime Day is real evidence. The year-over-year comparison depends on an undisclosed severance split, and two cost items are pending: Q4 fuel, which management expects to be higher, and the $1.5B-plus pay investment, which has no stated timing. Neither was sized for margin. The pillar stays at risk for one more quarter. The Q4 test is a clean year-over-year comparison. Q4 2025 North America margin was 9.0% as reported and about 9.5% excluding the $610M physical-store impairment charged mainly to that segment. A Q4 margin at or above that level, absorbing both cost items, would restore the pillar to on track.

10. Advertising: The First Miss in Seven Quarters

Management spent more prepared-remarks time on advertising than on any retail line, which signals that it expected the question. It quantified the Prime Day effect at 850 bps of sequential impact, named the NBA and search-marketing comparisons, and put the Q4 recovery in writing in the release. It also took on the bear thesis directly: external agentic shopping is not "meaningfully impacting Amazon Ads."

"First, we shifted Prime Day into Q2 for most countries, moving significant sales and ad revenue that were in Q3 2025 into Q2 2026, a tailwind to our Q2 2026 rate and a headwind to Q3 2026. For advertising, that was 850 basis points sequential impact to Q3 year-over-year growth." — Andy Jassy, CEO

New products were the support for the Q4 guide: Amazon Ads Agent, which runs the full advertising offering on one AI-first platform, and full-funnel campaigns, where advertisers set budget and duration and Amazon's systems plan and optimize the rest. In Q2, management said shoppers who click a sponsored prompt in its conversational experiences convert 48% more often. This quarter's Alexa statistic was that U.S. users of the assistant spend 50% more per order.

Assessment: The timing explanation is credible, and the agentic question was answered rather than avoided. But Amazon now carries a specific Q4 commitment on a line it does not normally guide. That raises the cost of a second soft quarter. Our Bull 4 status depends on the high teens being delivered.

11. Grocery, Speed, and the Retail Margin Claim

The retail narrative widened. Fresh groceries and Everyday Essentials grew about three times faster than the rest of the store and are about half of units. Grocery gross sales, including Everyday Essentials, are near $250B on a trailing basis. That is not comparable with the U.S. grocery figure of over $150B cited in Q2, which must have used a narrower scope. Amazon Now is in more than 450 towns and cities with item volume up 10 times, Pharmacy customers are up over 60%, and Prime Video is now profitable on a run rate "in the tens of billions."

"I'm not aware of another retail business approaching Amazon's revenue size, hundreds of billions of dollars, with double-digit operating margins inclusive of ads that's growing at this pace." — Andy Jassy, CEO

That claim does not match the reported segments. North America and International together earned $10,303M on $155,960M of revenue this quarter, a 6.6% margin, and 6.8% on a trailing basis. Both segments include advertising. A double-digit figure would require a basis that excludes something, most plausibly the costs of new businesses such as Leo, Zoox and video content that sit inside the segments. Amazon has never published that basis.

Assessment: The grocery data is the best evidence for the retail margin pillar, because basket size and frequency on a fixed same-day network is how density shows up in margin. The "double-digit" claim, by contrast, should not be repeated without the basis behind it. If Amazon wants the market to value retail on a margin structure that excludes its investment businesses, it needs to disclose those businesses' costs.

12. Below the Operating Line: The EPS Miss, the Mark That Moved to Equity, and Leo

Three items below or beside the operating line matter for modeling. First, other income was a $703M expense, against $53.4B of income in Q2 and $10.2B a year ago. No investment gain ran through the income statement this quarter. Second, the gain moved to equity: other comprehensive income included $35,974M of unrealized gains on available-for-sale debt securities, net of $12,233M of tax, about $48.2B pre-tax. Accumulated OCI rose $36.0B to $102.3B, and other assets rose $74.6B to $358.8B. The release explained the year-ago Anthropic gain and said nothing about the gain recorded in equity this quarter. Third, Leo capitalization was pushed out.

"Amazon Leo's commercial service is on track to begin initial service this year, and we expect to begin capitalizing certain costs in Q1 of next year, including production and launch costs. This is later than I've been predicting in earlier earnings calls. As a result, we will continue to expense the majority of Leo-related costs in Q4." — Brian Olsavsky, CFO

Assessment: The EPS miss tells us nothing about the operating business, and our ex-mark basis was built for exactly this. Leo capitalizing from Q1 2027 is a mechanical operating-income tailwind next year (costs move to the balance sheet and depreciate later) with no cash effect. It flatters FY27 margin, and we back it out when we judge the underlying trend. The $48.2B gain in equity is now the third consecutive quarter of a very large mark with no methodology or sensitivity disclosed.

Guidance & Outlook

Amazon landed Q3 revenue $56M below the top of its range and operating income $1.6B above the top, a 5.9% overshoot on the profit line after 14.4% in Q2. The Q4 guide looks soft against the Street on operating income, but it carries two named costs and a possible third. The majority of Leo costs stay in opex for one more quarter, and fuel and line-haul inflation is "a little bit higher than Q3." The release also announced a $1.5B-plus investment in U.S. operations pay, without saying when it starts or how much of it falls in Q4.

"Q4 net sales are expected to be between $240 billion and $247 billion. We estimate the year-over-year impact of changes in foreign exchange rates based on current rates, which we expect to be a headwind of approximately 60 basis points in the quarter. Q4 operating income is expected to be between $28.5 billion and $33.5 billion." — Brian Olsavsky, CFO

Implied quarter-over-quarter ramp: The revenue midpoint of $243.5B is a 20.6% sequential increase on Q3, against 18.4% in the same step last year. That reflects Q3's missing Prime Day, so the year-over-year rate is the better gauge. At the midpoint, Q4 grows 14.1% as reported and about 14.7% ex-FX, below Q3's roughly 17% underlying ex-FX rate. At the top of the range, Q4 grows 15.8% as reported and about 16.4% ex-FX, which matches Q3's underlying trend. The guide therefore assumes some deceleration unless Amazon lands at the top, as it did in Q3. The operating income midpoint implies a 12.7% margin, below Q3's 13.9%, consistent with Q4's seasonally lower-margin retail mix and the Leo expense. Q4 2025 operating income included $2.4B of special charges (tax disputes and a lawsuit settlement, severance, and physical-store impairments). Without them it was $27.4B, a 12.8% margin, so the guide midpoint is flat to slightly down on a clean basis rather than up 102 bps.

Street at: Consensus entering the print had Q4 revenue at $244.51B, $1.0B above the guide midpoint and inside the range. Operating income was near $33.0B to $33.2B, close to the $33.5B top. FY26 at the guide midpoints is $827.6B of revenue (+15.4%) and $110.4B of operating income at a 13.3% margin, against 11.2% in FY25.

Guidance style: Consistently conservative on profit. Amazon exceeded the midpoint of its operating income range by $5.5B in Q2 ($27.5B against $22.0B) and by $3.6B in Q3 ($28.1B against $24.5B), and cleared the top of the range both times. Landing at the top of the new $28.5B to $33.5B range would already meet the Street, and repeating Q3's $1.6B overshoot of the top would put Q4 near $35B. The disappointment against the Street is in the midpoint, not in what the pattern implies.

Analyst Q&A Highlights

Why the Sequential AWS Add Slowed

The first question went straight to the gap between accelerating year-over-year growth and a smaller sequential dollar increase, and asked whether customers, token pricing, or capacity explained it. Management answered at length on demand drivers (the AI stack, Trainium price-performance, Graviton for CPU-heavy AI work) and briefly on capacity, saying quarterly additions are uneven. It gave no Q4 capacity number.

Q: "Andy, continued to see the cloud business accelerate, which is really nice, but the quarter-over-quarter growth slowed. I am just wondering if there is any call-outs there on the customer side or token costs or anything that kind of drove that and how you are thinking about the capacity adds in Q4."
— Justin Post, Bank of America

A: "We add a lot of capacity every quarter. The amount that we add every quarter is not equal. We are adding different amounts each quarter. If I look at how much we have added this year and what we have planned over the next, I will call it, eight quarters or so, which is about as far out as we get eight to 10 quarters. We have a lot of capacity coming as well."
— Andy Jassy, CEO

Assessment: The answer identified capacity timing as the cause without saying so explicitly, and ruled out nothing on the customer or pricing side, because the demand list was the reply. That is consistent with the backlog. The specific request, Q4 capacity adds, was not answered for the second quarter running.

How AWS Gets to a Trillion Dollars

A question noted that a trillion-dollar statement was rare for this management team, more than it usually gives, and asked whether it was built top-down or bottom-up. The answer was mostly bottom-up for the near term and thematic beyond it. Management said big capacity blocks are reserved two to three years out, then listed new token-consuming use cases: security, coding, AI companions, and agentic rewrites of enterprise applications.

Q: "Can you just maybe walk us through how you get to that trillion of dollars? Is it top-down? Is it bottom-up? What gives you confidence that you are going to be able to put up that type of revenue growth and to get to that type of a number in the next five to 10 years?"
— Brian Nowak, Morgan Stanley

A: "We have very good visibility into 2027 and 2028 and some of 2029 too, with respect to what customers want in those really big chunks of capacity. ... You can't predict with any certainty five to 10 years from now, especially at these types of growth rates, but it's not a stretch for us to imagine that this could be over a trillion-dollar business."
— Andy Jassy, CEO

Assessment: The useful content is the visibility statement: 2027, 2028 and part of 2029 in the large capacity blocks. That is what supports the 40%-plus outlook. The trillion-dollar figure was explicitly hedged by management itself, which is the right way to read it.

Whether High-30s AWS Margins Hold for Two Years

A question asked directly whether the current AWS margin structure is stable over two years. The CFO declined to guide, defended the current margin as the product of discipline rather than chance, and then named the mix headwind. The second half of the question, on perishables coverage, drew strong engagement data but no coverage number.

Q: "Could I ask you to draw out on margins too? I know that margins can move up and down quarter-to-quarter, but over the next two years, do you think the current margin structure is relatively stable? We kind of hold at these kind of high 30%s, 40% operating margins for AWS."
— Mark Mahaney, Evercore ISI

A: "We are not guiding at all on our margin for AWS, but I can give you my color on it. ... In fact, the outsized growth of AI puts margin pressure on the business as a whole because the margins are lower there."
— Brian Olsavsky, CFO

Assessment: The refusal to guide is standard. Naming the mix headwind in the same answer is not. Management gave a revenue floor for 2027 and 2028 and, asked the matching margin question, pointed down. We expect AWS margin to drift lower from here.

Bedrock's Ceiling and the External Trainium Trade-Off

A two-part question asked what Amazon has learned about the Bedrock opportunity, and how it decides between selling Trainium externally and keeping chips for its own cloud. The Bedrock answer rested on model plurality: no single model wins, and customers want selection plus security. The chip answer was the more informative half, a clear statement of the accounting and cash trade-off, followed by the late-2027 start date.

Q: "Then with respect to the decision to sell Trainium externally, how do you think about balancing the market opportunity of external sales relative to the need for compute for that opportunity relative to other opportunities broadly across both internal use cases and AWS?"
— Eric Sheridan, Goldman Sachs

A: "If you sell externally, you are able to realize more revenue in the short term because you get to recognize the revenue as soon as it leaves your inventory. On the flip side, if you decide to use it inside your cloud in the form of EC2 instances or Bedrock, you actually will end up driving more absolute revenue and free cash flow from those chips because you get to use them over five to six years and drive revenue for a longer period of time."
— Andy Jassy, CEO

Assessment: By management's own logic, every chip sold outside earns less lifetime cash than one deployed inside, and the split will be reset each year inside a fixed foundry allocation. Management calls the external business "very significant," but that logic argues for a measured FY28 contribution in the model until volumes are disclosed.

The Capex Required for 40% Growth

The most direct cost question of the call asked what capital the 40%-plus outlook requires, and how management thinks about the payback period on servers, networking and data centers. The break-even part got a directional answer (management is "continuing to move that left") and the capex part got none. The CEO argued that the opportunity's timing is not Amazon's choice.

Q: "When you think about the 40%+ AWS revenue growth each of the next two years and then the multi-year path beyond that, can you just talk about the CapEx required to drive that kind of revenue over the next couple of years and beyond?"
— Doug Anmuth, J.P. Morgan

A: "Would it be more convenient if this type of acceleration and opportunity happened over 25 years and you could spread the CapEx over 25 years? Yeah, it might be more convenient, but that's not the world that we live in, and that's not the opportunity. We can fund this plan."
— Andy Jassy, CEO

Assessment: The question the stock most needed answered went unanswered, and the answer was clearly a choice. Last quarter management put a sub-three-year server break-even on the record. This quarter it said only that break-even is getting shorter, with no new number. The 2027 capex figure will now arrive with Q4 results, and the market will price the stock on it.

Bedrock's Margin as Inference Replaces Bare Metal

A question pointed out that most Bedrock token workloads run on Trainium and asked how the shift from selling infrastructure to selling inference as a service affects AWS revenue growth and margins. The answer confirmed the Trainium point and gave the most specific structural cost comparison of the call.

Q: "As this becomes a bigger part of the overall AWS story, like you just mentioned, Andy, how might that impact both revenue growth, but especially margins at AWS as you kind of shift from selling bare metal to selling more inference as a service?"
— Ross Sandler, Barclays

A: "Yeah. Well, you are right, that the majority and most of Bedrock today is being run on top of Trainium. We do that in meaningful part because we really like the price performance that we are getting, and it allows us to both keep prices as low as possible for customers, as well as our ability, we like the margins associated with it."
— Andy Jassy, CEO

Assessment: A clear confirmation that Bedrock and Trainium are one economic unit. It also partly offsets the CFO's mix warning: AI sold as tokens on owned silicon in owned buildings carries better margins than AI sold as rented GPUs. How the two forces net out is the AWS margin question for 2027.

Robotics Milestones for 2027

The second half of the same question asked what robotics milestones matter for retail in 2027. The answer gave no milestone. Instead it gave a fleet size (more than 1 million robots), a deployment rule (deploy where productivity, safety, cost and speed are comparable or better), and a statement that the impact so far is small compared with the next five years.

Q: "On the robotics side, as we look into 2027, what are the key milestones you are both focused on the robotics side for the Retail business?"
— Brian Nowak, Morgan Stanley

A: "If I look at the new designs that we have coming in our fulfillment network, including our expansion of same-day facilities, they are very heavy on robotics. We think it has had a big impact, and we think it is small compared to what it is going to be in the next five years."
— Andy Jassy, CEO

Assessment: No 2027 milestone was given, so there is nothing to grade next year beyond cost to serve per unit. The robotics case for the retail margin pillar remains a direction backed by a fleet count and one year of shipping-cost data, not a schedule.

What They're NOT Saying

1. The 2027 capex number, and 2027 depreciation: "Meaningfully more" than ~$220B, with no range, and for a third consecutive call no framing of what the 2026 capital vintage does to the depreciation line. D&A is already up 24.7% year over year to $20.9B a quarter. This is still the largest gap between the bull framing and the reported financials.
2. The chips and AI run rates: Both above $25B in Q2, both dropped this quarter in favor of growth descriptions and commitment multiples. Two quantified disclosures removed in one quarter, after the $50B standalone chips figure was removed in Q2, is a trend.
3. The Q4 capex reconciliation: The ~$220B plan implies about $47.8B of Q4 cash capex, 37% below Q3, in the quarter before a "meaningfully more" year. No one asked, and management did not volunteer how to reconcile it.
4. $63.8B of non-capex investment and a $48.2B unrealized gain: The investing line for acquisitions and non-marketable investments was $24.0B again this quarter and $63.8B year to date, and other comprehensive income carried about $48.2B of pre-tax unrealized gains on available-for-sale debt securities. Neither was mentioned on the call.
5. The cost of the wage investment: The release announced a $1.5B-plus investment to raise minimum starting pay for U.S. full-time operations employees to $20 an hour. It was not mentioned on the call, and neither its timing nor its effect on the Q4 guide or 2027 North America margin was given.
6. The basis for "double-digit" retail margins: The combined North America and International margin is 6.6% this quarter and 6.8% trailing, with advertising included. The basis behind the CEO's claim was not given.
7. Power capacity and the frontier model: Two Q2 commitments were not mentioned: doubling power capacity by end-2027 against 2025, and Amazon building its own frontier model. Neither was asked about, and neither was restated.
8. Perishables coverage: Asked directly how many Prime households have perishables checkout, management answered with engagement multiples and gave no coverage figure.
9. Energy derivatives in Q3: The Q4 guide "assumes no impact from energy derivative contract remeasurements," which contributed about $600M to Q2. Whether any landed in Q3's record margin was not said.

Market Reaction

• Pre-print setup: AMZN closed at $243.79 on October 29, down 4.2% in the regular session before the release, on a day when Alphabet fell 12.6% after its own report. The stock entered the print up 5.6% year to date against 12.8% for the S&P 500, up 5.9% over twelve months, and down 1.2% over 30 days. The 52-week closing range was $198.79 to $284.02, and the stock sat 14% below that high.
• After-hours move: Shares fell as much as about 7% in extended trading, to around $226.
• Reaction session (October 30, in progress at publication): The stock opened at $232.53, a 4.6% gap down, traded a $229.36 to $241.55 range, and was at $239.06 at midday, down 1.9%. Volume of 58.1M shares by midday was already 1.6 times the 30-day full-session average. At the same time the Nasdaq Composite was down 2.6% and the S&P 500 1.6%.
• Peer context: Alphabet fell 12.6% on October 29 after its own report, in a session that coverage tied to a broad technology sell-off. Microsoft was down 0.2% and Meta up 2.0% that day. At midday on the reaction session, Alphabet was down 1.0%, Microsoft 1.6%, Meta 0.6% and Oracle 5.1%.

A risk-off tape before the release. The print arrived at the end of a broad technology sell-off. Alphabet's decline in the October 29 session was its worst single-day percentage drop on record, the Nasdaq fell 1.4%, and Amazon fell 4.2% in the same session before releasing anything. The selling was not uniform across the large platforms, since Meta rose 2.0% that day. Into that tape, Amazon's own release added a ($42.4B) trailing free cash flow line and a "meaningfully more" 2027 capex statement.

The after-hours read was the consensus sheet; the session read was the call. The 7% after-hours decline priced the headline misses on EPS and revenue, the advertising shortfall, and the AWS gap to a buy-side bar in the low 40s. The regular session then recovered most of that: from an after-hours level near $226 and a session low of $229.36, the stock was at $239.06 at midday, while the Nasdaq was down 2.6%. On that intraday comparison, Amazon was outperforming the index on the day. The sequence suggests the market spent the morning on the 40%-plus outlook and the backlog, and decided they offset most of the headline damage.

Where it leaves the stock. At $239, Amazon is 12.0% below the $271.58 close after the Q2 print and 15.8% below the pre-print 52-week closing high. About 90% of the Q2 re-rating has been given back while the AWS outlook got stronger. The stock trades at about 25 times our FY27 ex-mark EPS estimate, against about 29.5 times on the same basis at the post-Q2 close.

Street Perspective

Debate: Was 39% AWS Growth a Miss?

Bull view: This was an expectations miss on a whisper number. AWS beat the published consensus on revenue and growth, accelerated for a sixth quarter, added the second-largest dollar amount in its history, and gave a two-year outlook above 40%. A business with a $578B backlog and capacity sold through 2028 does not have a demand problem.

Bear view: The sequential dollar add shrank by about $900M while capex rose 43% sequentially. Capital is going in faster than revenue is coming out, which is the definition of declining incremental returns. Until Amazon gives the capacity schedule, uneven capacity is an untestable excuse.

Our take: The bull is right on demand and the bear is right on disclosure. The backlog growth settles the demand question. But a capacity-gated growth rate needs a capacity schedule, and Amazon has refused to provide one for two quarters. Q4's dollar add settles it: $4B-plus confirms the timing explanation.

Debate: Can the Market Underwrite an Unsized 2027 Capex Bill?

Bull view: Management has a sub-three-year server break-even on the record, a backlog of 3.6 times trailing AWS revenue, and a 40%-plus revenue outlook that is itself the capex justification. Operating cash flow grew 29%, the balance sheet holds $613.6B of equity, and net debt is under $60B.

Bear view: Trailing free cash flow is ($42.4B) and heading lower. Year-to-date capex and non-marketable investments of $235.9B are double operating cash flow, long-term debt is more than 2.6 times its December level, and 2027 is "meaningfully more" with no number. In a market already selling technology on the day, a free cash flow deficit that widens again in 2027 is hard to look past.

Our take: The bear has the near-term argument and the bull has the economics. FY27 free cash flow will likely be worse than FY26 unless AWS revenue conversion accelerates faster than capex. That is already in our model and, at 25 times earnings, in the price. The risk is a 2027 number well above $270B, which would push the trough into 2028.

Debate: Is the Advertising Slowdown Timing or the Start of Agentic Disruption?

Bull view: Management sized the Prime Day effect at 850 bps and named the NBA and search-marketing comparisons, which together explain most of the drop from 26% to 12%. Q4 is guided to the high teens in writing, NBA returns, and the new advertising agent and full-funnel products are just launching.

Bear view: This was the first miss in seven quarters, and Amazon itself raised the agentic question in its prepared remarks. Several timing explanations at once is how a structural slowdown first looks. The search-marketing lap also means some of the prior acceleration was bought with traffic.

Our take: Timing, on the evidence available, and Q4 will confirm or refute it quickly. The explanation is quantified, the recovery is guided, and the agentic denial came with a reason (external agents are not yet more convenient). We would change our view on a Q4 print below 15%.

Debate: Is This the Entry Point?

Bull view: The stock has given back most of the Q2 re-rating while the AWS outlook strengthened, retail margin improved, and Trainium got an external launch date. At about 25 times FY27 ex-mark earnings for a company guiding its most profitable segment to 40%-plus growth, the multiple is well below where the stock traded after Q2.

Bear view: Earnings multiples understate capital intensity when free cash flow is deeply negative and depreciation is about to step up. The relevant valuation is on free cash flow, which is negative for at least another year, and the catalyst (the 2027 capex number) is more likely to disappoint than to reassure.

Our take: Closer to the bull. Free cash flow is the honest bear metric. But the backlog converts into revenue at 38%-plus segment margins, and the market is pricing the cost of the buildout without the return on it. The February capex number is the event risk, and it is why we are not raising the rating's conviction.

Model Update Needed

Valuation impact: Fair value moves to $285 to $315 from $295 to $315. We raise FY27 ex-mark EPS to about $9.50 from $9.20 on the 40%-plus AWS outlook, net of higher interest expense and a lower AWS margin. We lower the multiple to 30 to 33 times from 32 to 34 times, for 2027 capital intensity that is now explicitly higher. Against the $239.06 midday price, that is 19% to 32% upside. Even the bottom of the range implies a 12-month return well above what we expect from the S&P 500, which is the basis for Outperform rather than Hold. The bull case, with the 2027 outlook delivered, AWS margin held near 38% and retail margin at 7.5%, gives FY27 ex-mark EPS near $10.20; at 34 times that supports about $345, 44% upside, before any FY28 contribution from external Trainium. The bear case, where 2027 depreciation compresses consolidated margin toward 12.5% and ex-mark EPS falls to about $8.30 at 27 times, gives $224, 6% downside. The asymmetry is better than it was after Q2, because the price fell 12% while the operating outlook improved.

Thesis Scorecard Post-Earnings

Overall: Thesis unchanged in structure and stronger on its central pillar. AWS now has a two-year outlook that turns the reacceleration thesis from a trend into a management commitment, and the silicon pillar has a date. The retail margin pillar passed the test we set for it, though not cleanly enough to restore. Advertising had its first soft quarter, with a quantified explanation and a written recovery target. The free cash flow bear is fully materialized and growing, and the depreciation bear is unresolved for a third quarter. No status tag moved. The scorecard now hinges on two numbers that arrive with Q4 results: the 2027 capex figure and the Q4 advertising growth rate.

Action: Maintaining Outperform, with conviction unchanged. At $239, the stock has given back most of the Q2 re-rating while the AWS outlook improved, and our $285 to $315 range implies 19% to 32% upside. Our downgrade triggers are updated. The North America trigger was tested and passed, and it is replaced by a Q4 test: a clean-basis year-over-year decline, which means a Q4 North America margin below about 9.5%, Q4 2025's margin excluding its impairment charge. The depreciation trigger stands. We add two more: Q4 advertising growth below 15%, or a 2027 capex figure above roughly $300B without a matching raise in the AWS outlook.