AMZN · Q3 2025 Earnings

Bullish

Amazon

Reported October 30, 2025

30-second summary

AWS grew 20% YoY to $33.0B — the fastest pace since 2022 — confirming last quarter's "demand exceeds supply" framing was not a deflection. Q3 revenue of $180.2B (+13% YoY) beat the high end of guidance by $0.7B, and Q4 operating income guidance of $21.0–26.0B implies a $23.5B midpoint, roughly +11% YoY against the $21.2B prior-year base. GAAP operating income of $17.4B included $4.3B of special charges ($2.5B FTC settlement + $1.8B severance); ex-charges, operating income would have been $21.7B, which management framed as $1.2B above the high end of the guidance range. The combination of AWS reacceleration, $1B+ ARR in agentic AI (Connect), and 150% QoQ Tranium 2 revenue growth tells you the AI infrastructure spend is converting to revenue faster than the bear case allowed.

Headline numbers

EPS

Q3 FY2025

$1.95

Revenue

Q3 FY2025

$180.20B

+13.0% YoY

Gross margin

Q3 FY2025

50.8%

Operating margin

Q3 FY2025

9.7%

Key financials

Q3 FY2025

Metric	Q3 FY2025	YoY	Q2 FY2025	QoQ
Revenue	$180.20B	+13.0%	$167.70B	+7.5%
EPS	$1.95	—	$1.68	+16.1%
Gross margin	50.8%	—	51.8%	-100bps
Operating margin	9.7%	—	11.4%	-170bps

Guidance

Amazon beat Q3 revenue guidance, raising Q4 operating income outlook to $21–$26B while reaffirming 10–13% revenue growth; AWS re-accelerating at 20.2% YoY signals strong AI momentum.

Guidance is issued one quarter forward. The Prior-guide column references the guide issued last quarter for the period just reported; the New-guide column is for next quarter.

Actuals vs prior guidance

Metric	Period	Prior guide	Actual	Δ	Result
Revenue	Q3 FY2025	$174.0 billion to $179.5 billion	$180.2 billion	+$0.7 billion above guide high	Beat
Revenue YoY growth	Q3 FY2025	10% to 13%	13%	+0 pts at top of guide	Beat
Operating Income	Q3 FY2025	$15.5 billion to $20.5 billion	$17.5 billion	-$3.0 billion below guide midpoint	Beat

New guidance

Metric	Period	Guide	YoY
Revenue	Q4 FY2025	$206.0 billion to $213.0 billion	10% to 13%
Revenue YoY growth	Q4 FY2025	10% to 13%	—
Operating Income	Q4 FY2025	$21.0 billion to $26.0 billion	—
FX impact	Q4 FY2025	approximately 190 basis points favorable	—

Segment performance

Q3 FY2025

Segment	Q3 FY2025	YoY
North America	$106.3B	+11.0%
International	$40.9B	+14.0%
AWS	$33B	+20.0%
Online Stores	$67.4B	+10.0%
Third-party Seller Services	$42.5B	+12.0%
Advertising Services	$17.7B	+24.0%
Subscription Services	$12.6B	+11.0%

Platform metrics

Q3 FY2025

Segment	Q3 FY2025
AWS YoY Growth	20.0%
Advertising Services YoY Growth	24.0%
Third-party Seller Services YoY Growth	12.0%
Worldwide Paid Units YoY Growth	11.0%
Seller Unit Mix	62%
Connect AI Solution ARR	$1.0 billion

Profitability

Q3 FY2025

Segment	Q3 FY2025
AWS Operating Margin	34.6%
Operating Cash Flow YoY Growth (TTM)	16.0%

Management tone

Narrative arc: Q2 "capacity is the constraint, demand is there" → Q3 "capacity converted, AWS reaccelerated, AI is generating discrete revenue lines."

AI shifted from forward narrative to current revenue line. A quarter ago, Jassy was rebutting share-loss concerns by citing backlog ($195B, +25% YoY) and the long migration tail (85–90% IT spend still on-prem). This quarter, the rebuttal is the print itself: AWS growing 20.2% "at a pace we haven't seen since 2022," Tranium 2 revenue up 150% QoQ, Connect's agentic AI at $1B ARR. The framing collapsed from "trust the demand signal" to "look at the revenue."

Capacity language pivoted from constraint to expansion plan. Q2's headline was "power is the single biggest constraint" and "supply easing will take several quarters." Q3 puts numbers behind the build: 3.8 gigawatts added in the past 12 months, another gigawatt-plus in Q4, capacity to double by end of 2027. Project Rainier scaling from 500,000 to 1 million Tranium-2 chips by year-end. Power is still cited as the bottleneck, but the curve management is asking investors to anchor on is the multi-year doubling, not the quarterly throttle.

Headcount restructuring framed explicitly as NOT primarily AI-driven. Mahaney pressed on whether the ~18,000 role reduction was an AI efficiency move. Jassy's answer — culture/structure-driven, removing layers, increasing ownership — is notable for what it doesn't claim. Management is reserving AI productivity savings as future optionality, not current run-rate. "No commitment to flat headcount going forward" leaves room for both directions.

Q&A highlights

Justin Post · Bank of America

How capacity constrained is AWS, and what demand are you seeing for Tranium chips outside of major customers?

AWS has added 3.8 gigawatts in the past 12 months with another gigawatt-plus in Q4, expecting to double capacity by end of 2027. Power is currently the bottleneck. Tranium 2 is fully subscribed with 150% QoQ revenue growth and is a multi-billion dollar business. Tranium 3 should preview end of year with broader customer adoption expected, including medium-sized customers.

3.8 gigawatts added in past 12 monthsAt least 1 gigawatt expected in Q4 2025Capacity to double by end of 2027Tranium 2 revenue grew 150% quarter-over-quarter

Brian Nowak · Morgan Stanley

What are the key hurdles Tranium 3 must overcome to drive broader adoption versus reliance on third-party chips?

AWS will maintain multiple chip options including deep NVIDIA partnership. Tranium 3 expected to offer ~40% better price performance than Tranium 2 (which already has significant advantage). Success requires: chip delivery, volume production, rapid timeline, and strong software ecosystem. Proof points like Project Rainier build credibility and customer confidence.

Tranium 3 expected to be ~40% better than Tranium 2Tranium 2 already offers significant price performance advantageAWS buys significant NVIDIA volumes, not constrainedGraviton CPUs deliver ~40% better price performance versus x86

Mark Mahaney · Evercore ISI

Has Amazon reached scale in perishables delivery where it can reshape grocery habits? And how will headcount evolve given AI efficiencies?

Amazon's grocery business (excluding Whole Foods and Fresh) exceeds $100B in GMV annually. Same-day perishables delivery from 1,000 cities expanding to 2,300 by year-end is changing grocery shopping patterns. On headcount: Recent restructuring (~18,000 roles eliminated) was culture/structure-driven to remove layers, increase ownership, and operate lean—not primarily AI or financially motivated. No commitment to flat headcount going forward.

Grocery business (excl. Whole Foods/Fresh) >$100B GMV in past 12 monthsSame-day perishables delivery in 1,000 cities, expanding to 2,300 by year-endPerishable shoppers return 2x more often than non-perishable shoppers~18,000 roles eliminated to reduce organizational layers

Doug Anmuth · JP Morgan

What differentiates Project Rainier architecture and how will it be replicated with Tranium 3 for other customers?

Project Rainier's differentiation lies in Tranium 2 chip performance and price performance, combined with AWS's large-scale infrastructure capabilities to operate a 500,000+ chip cluster. The infrastructure feat of building such scale clusters is difficult. Rainier is Anthropic-specific, but AWS has other customers interested in large Tranium clusters for Tranium 3.

Project Rainier contains ~500,000 Tranium-2 chipsAnthropic building and deploying Claude on Tranium-2Expanding to 1 million Tranium-2 chips by year-endInfrastructure capabilities for large-scale cluster management unusual in industry

Eric Sheridan · Goldman Sachs

How will robotics and automation drive efficiencies and reinvestment potential across physical operations?

AWS has over 1 million robots in fulfillment network with significant ongoing invention expected. Robotics improve safety, boost productivity, increase speed, and allow humans to focus on problem-solving. Expectation is complementary human-robot fulfillment network over time. Continued significant investment expected to improve cost structure and customer experience.

Over 1 million robots currently in fulfillment networkRobotics improve safety, productivity, and speedHumans remain centerpiece of fulfillment networkExpected complementary human-robot operations model long-term

Answers to last quarter's watch list

AWS revenue growth versus 17.5% — AWS printed 20.2%, a 270bps sequential reacceleration, clearing the 18% threshold set as evidence the supply story is real. Tranium 2 +150% QoQ and capacity doubling target by 2027 corroborate that backlog is converting.

Resolved positively

Q3 operating income inside the $15.5–20.5B band — Landed at $17.4B GAAP, weighed down by $4.3B of special charges ($2.5B FTC settlement + $1.8B severance). Ex-charges, OI would have been $21.7B — $1.2B above the high end of the guidance band per the CFO. Underlying operating performance beat; the GAAP print understates the run-rate. Q4 guide midpoint of $23.5B signals leverage returns.

Resolved positively

Tariff impact on retail unit economics — North America operating margin printed 4.5% GAAP (6.9% ex-FTC charge), but worldwide paid units grew 11% and seller mix held at 62%, with no qualitative tariff escalation language in the prepared remarks. Management didn't call out a tariff-driven ASP move.

Continue monitoring

International margin trajectory — International revenue grew 14% to $40.9B; segment operating margin was 2.9% (includes a portion of the severance charge). Ex-severance, management noted international operating margin expanded year over year, but the precise magnitude isn't isolated.

Continue monitoring

Alexa Plus and Kuiper monetization signals — Alexa Plus engagement metrics disclosed (2x interaction rate vs. classic Alexa, 4x shopping conversations ending in purchase) but no paid subscriber count; Kuiper fleet expanded to 150+ satellites with commercial agreements announced (JetBlue, Australia NBN, Kazakhtelecom) but no firm revenue contribution. Both remain optionality.

Continue monitoring

What to watch into next quarter

AWS growth above 20% — does the reacceleration continue into Q4, or was 20.2% a one-quarter step-up as capacity came online? A print of 21%+ would confirm the doubling-by-2027 capacity build is feeding revenue faster than expected; a deceleration back toward 18% would suggest Q3 was a release of pent-up backlog rather than a new run rate.

Q4 operating income relative to $23.5B midpoint — the upper half ($23.5–26.0B) would confirm AI capex is producing margin leverage; the lower half would signal Q3's underlying margin compression extends into the seasonal peak.

Tranium 3 customer disclosures — management previewed Tranium 3 for year-end with "broader customer adoption including medium-sized customers." Watch for named customer wins beyond Anthropic on the Q4 print as evidence the chip strategy scales beyond hyperscaler-tier accounts.

Connect ARR trajectory beyond $1B — first cross-quarter data point on agentic AI revenue. A print of $1.5B+ on Q4 would establish a credible growth curve; flat-to-slow would suggest the $1B was front-loaded enterprise pilots.

Headcount direction post-restructuring — Jassy explicitly refused to commit to flat headcount. Whether Q4 shows further role reductions or hiring resumption will clarify whether the ~18,000 cut was a one-time structural reset or the start of an AI-driven efficiency program management isn't yet ready to label.

Sources

Amazon Q3 FY2025 earnings press release (SEC filing): https://www.sec.gov/Archives/edgar/data/1018724/000101872425000121/amzn-20250930xex991.htm
Q3 FY2025 earnings call Q&A.
Tapebrief Q2 FY2025 AMZN brief (for cross-quarter context).

Get the next brief, free.

We publish analyst-grade earnings briefs the same day or morning after every call — headline numbers, segment KPIs, Q&A highlights, and tone analysis. Free during beta.

This is not investment advice.