AWS Accelerates Again to +24% — But $200B 2026 Capex Resets the FCF Curve and the Market Spooks; Multiple Compresses Into a Stronger Print
Key Takeaways
- AWS accelerated again — to +24% YoY at $35.6B — the highest growth rate in 13 quarters and the second consecutive sequential acceleration after Q3's +20.2%. Backlog stepped up to $244B (+40% YoY, +22% QoQ from $200B Q3) — biggest sequential backlog jump in history. Custom silicon (Graviton + Trainium) crossed $10B+ annual run rate growing triple-digit YoY. Trainium2 has 1.4M chips landed (fastest ramping chip launch ever); Trainium3 is launched and "nearly all supply committed by mid-2026."
- The capex headline: $200B planned for 2026 — a +56% step-up from FY25's $128.3B and the largest single-year capex commitment by any company in history. The framing was unequivocal: "we are monetizing capacity as fast as we can install it." The market reacted with an ~8–11% after-hours selloff (subsequently extending into next-day trading) — the biggest one-day move on AMZN earnings since 2022 — pricing 2026 FCF compression risk despite the operational beat.
- Underlying retail and ads were excellent: NA $127.1B (+10%) at 9.0% margin (+100bps YoY, a record), advertising $21.3B (+22% — fourth consecutive accelerating quarter; Prime Video ad-supported audience 315M vs. 200M early 2024), worldwide paid units +12% (highest of FY25). "Add to delivery" already at ~10% of Prime volume 6 months post-launch. Bedrock customer spend +60% QoQ. Rufus reached 300M+ users in 2025.
- Q1 2026 op-income guide $16.5–$21.5B is below pre-print Street ~$22.0B at midpoint, weighed by an explicit ~$1B YoY NA cost increase tied to Amazon LEO satellite launches (commercial launch 2026; AT&T, DirecTV Latin America, JetBlue, Australia NBN signed). Full-year 2026 op-income outlook unstated but the Q1 guide and capex curve mean FY26 FCF will compress further from FY25's already-pressured $11.2B.
- Rating: Maintaining Outperform. The operational thesis is the strongest it has been at any point in the four-quarter window — AWS reaccelerating twice in a row, custom silicon scaling faster than the market models, advertising compounding into a structural CTV moat, retail margin at a Q4 record. The selloff reduces the multiple by 8–11% on a print where forward earnings power expanded — that is the textbook setup for an Outperform call. We are buyers into the gap.
Results vs. Consensus
| Metric | Actual | Consensus | Beat/Miss | Magnitude |
|---|---|---|---|---|
| Revenue | $213.4B | ~$211.4B | Beat | +0.9% |
| EPS (GAAP, diluted) | $1.95 | ~$1.97 | Miss | −1% |
| Operating income (reported) | $25.0B | ~$23.5B | Beat | +6% |
| Operating income (ex-charges) | $27.4B | ~$23.5B | Beat | +17% |
| AWS revenue | $35.6B | ~$34.0B | Beat | +4.7% |
| AWS growth rate | +24% | ~+20% | Beat | +400bps |
| Advertising revenue | $21.3B | ~$20.0B | Beat | +6.5% |
| NA op margin | 9.0% | ~7.5% | Beat | +150bps |
| Q1'26 op-inc guide (mid) | $19.0B | ~$22.0B | Miss | −14% |
| 2026 capex | ~$200B | ~$145B | Above | +38% |
Quality of Beat/Miss
- The AWS beat is exceptional and high quality: +24% on $142B run rate is mathematically harder than +35% on a $40B run rate. Sequential revenue +$2.6B, +$7B YoY. Backlog +$44B sequentially is the largest absolute backlog step-up ever. This is not noise; it is durable acceleration.
- EPS slightly missed on the special charges ($2.4B aggregate) — clean EPS ex-charges would have been ~$2.10–$2.15, well above consensus.
- Op-income beat ex-charges of $27.4B is +$3.9B above Street midpoint — driven by AWS reacceleration plus NA margin record (9.0%, +100bps YoY).
- Advertising +22% in Q4 is the fourth consecutive accelerating quarter (+18% Q1, +22% Q2, +22% Q3, +22% Q4 — note Q3 and Q4 both at 22% but on accelerating absolute base; FY25 added $12B+ incremental).
- The "miss" is forward-looking: Q1'26 guide midpoint is $3B below Street, and 2026 capex of $200B is $55B above Street. Both are management telling us the business is investing harder, not earning less.
Segment Performance
| Segment | Revenue | YoY (ex-FX) | Op Income | Op Margin | Notable |
|---|---|---|---|---|---|
| North America | $127.1B | +10% | $11.5B | 9.0% (+100bps YoY) | Record Q4 margin; $610M asset impairment + portion of $730M severance |
| International | $50.7B | +11% | $1.0B | 2.1% | $1.1B Italy tax/lawsuit charge; ex-charges margin expanded YoY |
| AWS | $35.6B | +24% | $12.5B | 35.0% (+40bps YoY) | $142B run rate; backlog $244B (+40% YoY); 1+ GW added Q4 |
AWS — Acceleration Sustained, Backlog Step-Up Confirms 2026 Visibility
AWS revenue of $35.6B was +24% YoY — the second consecutive sequential acceleration (Q2: +17.5%, Q3: +20.2%, Q4: +24%). On a $142B run rate, this is the strongest absolute-dollar growth of any cloud business in any quarter ever. Sequential revenue added $2.6B; YoY add was $7B. The 2025 capacity addition (more than any other company) is now visibly converting to revenue at scale. AWS operating income of $12.5B at 35% margin is +40bps YoY — confounding the bear narrative that the 2026 capex curve necessarily compresses margin.
"AWS growth continued to accelerate to 24%, the fastest we've seen in thirteen quarters. Up $2.6 billion quarter over quarter and nearly $7 billion year over year. AWS is now a $142 billion annualized run rate business." — Andy Jassy, CEO
The backlog disclosure is the structural story. $244B at year-end (+40% YoY, +22% sequentially) — a $44B sequential step-up. That is the largest absolute backlog increase in AWS history and provides clear FY26 revenue visibility. At 4.1-year duration, this is roughly $60B of annualized visible revenue still to convert.
The customer wins list is the most aggressive in any quarter since AWS launched: OpenAI (notably), Visa, NBA, BlackRock, Perplexity, Lyft, United Airlines, DoorDash, Salesforce, US Air Force, Adobe, Thomson Reuters, AT&T, S&P Global, NBC Sports, LSE, Choice Hotels, Accenture, Indeed, HSBC, CrowdStrike. The OpenAI mention is the most consequential — a partnership extension that expands AWS's role in frontier-model training beyond the Anthropic relationship. Top US 500 startups: more use AWS as primary cloud than the next 2 providers combined.
Assessment: The supply-side bottleneck has been substantially relieved on chips; power remains the industry-wide constraint but AWS is leading the industry in adding it. AWS is now in a growth phase that the market under-modeled into the print, and Q4 confirms the Q3 reacceleration was structural, not one-off.
Custom Silicon — From Optionality to $10B+ Run Rate
The single most important new disclosure: Graviton + Trainium combined are now a $10B+ annual run rate business growing triple-digit YoY. Trainium2 alone has 1.4M chips landed — the fastest-ramping chip launch in AWS history. Multibillion-dollar run rate, 100K+ companies using it, and the "majority underpinning of Bedrock usage today." Trainium3 is now launched, claimed at ~40% better price/performance vs. Trainium2, and "nearly all supply committed by mid-2026." Trainium4 in build with "very substantial interest already" (2027 delivery). Trainium5 already in conversations.
"I think people know about our chips capability, our chips business, but I'm not sure folks realize how strong a chips company we've become over the last ten years." — Andy Jassy, CEO
Graviton (CPU) is at +50% YoY with 90%+ of AWS top 1,000 customers using it. Trainium (AI accelerator) is fully subscribed across both v2 and v3 supply. The strategic implication: AWS now captures both the cloud-compute revenue and the silicon margin layer, structurally compounding gross margin over the next 24 months as the silicon mix shifts to custom from NVIDIA-only.
Assessment: Custom silicon is the AWS AI gross-margin moat. The bear case that Trainium is single-customer (Anthropic) is retired by the 100K+ company adoption. Trainium3 supply being committed by mid-2026 means visibility extends through year-end.
$200B 2026 Capex — The Number That Spooked the Market
The headline that drove the after-hours selloff: 2026 capex guided to ~$200B. That's +56% from FY25's $128.3B (which itself was +65% from FY24's $77.7B). For context, $200B in a single year is the largest capex commitment of any company in history — exceeding the most aggressive Microsoft / Google / Meta data-center estimates. Olsavsky and Jassy both framed it consistently: predominantly AWS, predominantly AI, monetized as installed.
"We expect to invest about $200 billion in capital expenditures across Amazon... predominantly in AWS, because we have very high demand. We are monetizing capacity as fast as we can install it. We have deep experience understanding demand signals in the AWS business and then turning that capacity into strong return on invested capital." — Andy Jassy, CEO
Mark Mahaney's question — about long-term ROIC and FCF guardrails — was the question of the call. Jassy and Olsavsky declined to commit to specific FCF floors but layered the case: capacity converts immediately, AWS margins are durable, custom silicon improves the unit economics, and the ROIC pattern from the original AWS buildout will repeat. Mathematically: if AWS LTV per dollar of capacity holds at the recent run rate, $200B converts to ~$50B of incremental annualized AWS revenue at ~35% margins — meaning ~$17B of incremental AWS operating income contribution from FY26 capex landings flowing through FY27/28.
Assessment: The $200B number is shocking because of magnitude, not because of math. The AWS operational data points (24% growth, $244B backlog, custom silicon at $10B run rate) all support the underlying economics. The FCF compression is real and visible — TTM FCF of $11.2B is the lowest since 2022 — but the LTV math justifies it. The market reaction is rational on multiple compression but priced into a stronger forward earnings curve.
North America — Record Q4 Margin
NA revenue $127.1B (+10%) with operating income $11.5B at 9.0% margin (+100bps YoY) — a record Q4 margin print. Adjusted for the $610M asset impairment (physical stores) and portion of severance, underlying NA margin would have been ~9.5%. Drivers: regionalized network at scale, U.S. inbound lead time −4 days, record delivery speeds (8B+ items same/next day, +30% YoY; same-day +nearly 70% YoY in U.S.), and operating leverage from advertising. Same-day perishables in 2,300+ U.S. cities; perishable shoppers shop 2× more often and add 3× more items per order.
Assessment: NA margin is now structurally above 8.5%, with a clear path to 10%+ as same-day perishables mature, "Add to delivery" volume scales, and 3P unit-mix and ad leverage continue to compound. This is the under-appreciated margin story.
International — One-Time Charges Mask Margin Progress
International $50.7B (+11% ex-FX) at 2.1% margin. The $1.1B Italy tax-and-lawsuit charge depresses the reported margin by ~220bps; ex-charges, International margin expanded YoY in line with the 11-quarter pattern.
Key KPIs
| KPI | Q4 2025 | Q3 2025 | YoY | Trend | Read |
|---|---|---|---|---|---|
| AWS revenue growth | +24% | +20.2% | — | Accel +380bps | 13-quarter high |
| AWS backlog (RPO) | $244B | $200B | +40% | Step-up | +22% QoQ — historic |
| AWS operating margin | 35.0% | 34.6% | +40bps | Up | Despite capex ramp |
| Custom silicon run rate | $10B+ | — | +100%+ | Accel | Graviton + Trainium combined |
| Trainium2 chips landed | 1.4M | — | — | Fastest ever | Trainium3 supply 50%+ committed |
| Bedrock customer spend | +60% QoQ | — | — | Accel | Multibillion-$ run rate |
| Advertising revenue | $21.3B | $17.7B | +22% | 4th accel | Prime Video 315M global audience |
| Worldwide paid units | +12% | +11% | — | Highest 2025 | Holiday + same-day |
| NA op margin | 9.0% | 4.5% / 6.9% ex-FTC | +100bps | Record | Network productivity + ads |
| Cash capex (Q4) | ~$38B | $34.2B | — | Up | FY25 total $128.3B |
| FCF (TTM) | $11.2B | $14.8B | −71% | Compress | Capex absorbing |
| 2026 capex guide | ~$200B | — | +56% | Step-up | Largest single-year capex ever |
Key Topics & Management Commentary
Overall Management Tone: Confident, expansive, and at times almost emboldened — the stark contrast with Q1's careful and Q2's guarded tone is the clearest evidence of how dramatically the AWS narrative has changed. Jassy spent the bulk of the call on AI strategy depth (NovaForge, Trainium roadmap through 2027/28, Frontier Agents). Olsavsky was unusually direct on capex framing — "long arc of additional revenue we see from other customers and backlog and commitments" — and declined to commit to FCF floors when explicitly asked. The implication: management is willing to absorb FCF compression for 12–24 months in exchange for capturing the AI workload migration.
The $200B Capex Decision
The most important decision Amazon has communicated in any call we have analyzed. $200B is not a number that materializes from a single demand signal — it represents power-purchase agreements, multi-year data-center construction commitments, custom silicon supply contracts (Trainium2/3/4), and NVIDIA orders. The pre-commitments imply confidence beyond Q4 backlog visibility into 2027–2028 demand.
"We are putting into service with customers all capacity that we are getting, and it's immediately useful. And we are also seeing a long arc of additional revenue we see from other customers and backlog and commitments... so you can see that it's working out its way into our P&L both through CapEx and also through our operating margin in AWS." — Brian Olsavsky, CFO
Mark Mahaney's question was the cleanest expression of investor concern: where are the FCF guardrails? Both Jassy and Olsavsky declined to commit to floors but framed the answer through the AWS LTV lens. The implicit message: 2026 will be a multi-year-low FCF year; FY27 will inflect; FY28 will be the durable normalization point.
Assessment: The $200B number is justifiable on the unit-economics math we walked through earlier. The risk is timing — if AWS demand surprises to the downside in any quarter, the multiple gets repriced sharply because capex commitments are largely fixed. We see this as the principal risk for 2026.
NovaForge — A Genuinely New Capability
One of the more under-appreciated launches: NovaForge gives customers early checkpoints on Amazon Nova models so they can mix proprietary data into the pre-training stage rather than only fine-tuning post-train. Jassy framed it as "teaching a child a foreign language early in their life." This addresses a real and growing enterprise pain point — the limits of post-train fine-tuning on competitive differentiation. "There is nothing else out there like this today" — and even with sales hyperbole discount, this is plausibly true given pre-training-checkpoint access has not been a commercially available offering at scale.
Assessment: Could be a significant differentiation in enterprise model adoption. We will track adoption metrics as they become available.
Frontier Agents — Autonomous Agents at Scale
At AWS re:Invent (December 2025), Amazon launched Frontier Agents — autonomous agents that "run persistently for hours or days, scale out quickly, and remember context." Specific products: Kiro autonomous (coding tasks beyond IDE assistance), AWS DevOps Agents (operational issue detection/resolution), AWS Security Agents (proactive securing through dev lifecycle). Kiro developer count grew +150% QoQ. AgentCore continues to expand. Strands (open-source agent framework) and AWS Transform (mainframe / VMware / .NET migration) are scaling fast — saved 700K+ hours of manual effort YTD.
Assessment: The agent-economy narrative is converting from concept to commercial. Kiro's +150% QoQ is the leading indicator. We expect "agent-related AWS revenue" to be a disclosed line item in FY27.
Retail Margin and Same-Day Perishables — The Quiet Compounding Story
NA at 9.0% Q4 margin, +100bps YoY, is the operational headline most analysts under-emphasize relative to the AWS noise. Drivers: regionalized network at scale, US inbound lead time −4 days, record delivery speeds, "Add to delivery" at ~10% of Prime volume in 6 months. Same-day perishables in 2,300+ U.S. cities; perishable shoppers shop 2× more often and add 3× more items per order. Amazon Now ultrafast (30-min) delivery launched in India / Mexico / UAE — Indian users tripled shopping frequency upon adoption.
Whole Foods adding 100+ stores over next few years. Daily Shop urban small-format expanding. Amazon Haul (sub-$10): 1M+ items in 25 countries. 150M Americans use Amazon as a "go-to grocery destination" with $150B+ grocery gross sales.
Assessment: The grocery / quick-commerce flywheel is the single largest under-modeled long-term TAM expansion. Amazon Now's Indian early-data point (3× frequency) is a serious signal.
Advertising — CTV Aggregator Status Confirmed
$21.3B (+22%) in Q4, fourth consecutive accelerating quarter. FY25 added $12B+ incremental ad revenue. Prime Video ad-supported audience: 315M global (vs. 200M early 2024). Available in 16 countries. Thursday Night Football: most-watched season ever, 15M+ avg viewers (+16% YoY); Packers vs Bears wild card was the most-streamed NFL game ever (31.6M viewers). Ads agent and creative agent (concept-to-completion AI campaign generation) launched.
Assessment: Amazon Ads is now structurally the #3 digital ad platform globally and the #1 CTV ad platform. The compounding here may be the single most durable line in the model.
Amazon LEO — Commercial Launch in 2026
180 satellites launched (formerly Project Kuiper). 20+ launches in 2026, 30+ in 2027. Commercial launch later in 2026. LEO Ultra terminal: up to 1 Gbps download / 400 Mbps upload — fastest satellite Internet antenna ever built. Customers signed: AT&T, DirecTV Latin America, JetBlue, Australia NBN. The Q1 2026 op-income guide includes ~$1B YoY NA cost increase tied to LEO launches; later in year many of these costs will be capitalized.
Assessment: LEO is no longer optionality — it's a 2026 commercial business. Enterprise/government anchor customers signed; AWS-pairing differentiation is becoming material. Costs are visible; revenue is forward-loaded.
Guidance & Outlook
| Metric | Q1 2026 Guide | Implied YoY | Street pre-print | Read |
|---|---|---|---|---|
| Net sales | $173.5B–$178.5B | +7% to +10% | ~$176B | Midpoint in line |
| Operating income | $16.5B–$21.5B | −10% to +17% | ~$22.0B | Midpoint below |
| FX impact | ~+180bps favorable | — | — | Tailwind |
| 2026 capex | ~$200B | +56% | ~$145B | Materially above |
Q1 op-income guide $16.5–$21.5B (midpoint $19.0B) is ~$3B below pre-print Street ~$22B. Olsavsky decomposed: +$1B YoY NA cost from Amazon LEO (most expensed; capitalization later in year), continued investment in International stores (delivery speed via Amazon Now, sharp pricing/seller-fee competition in select markets). Both of these are forward investments management is willing to make against returns that compound beyond FY26.
Implied Q1 op-income trajectory: midpoint $19.0B is +5% YoY (Q1 2025 was $18.4B). High end $21.5B is +17%. Low end $16.5B is −10%. Wide $5B band continues the four-quarter pattern.
Street at: ~$22.0B pre-print; will recalibrate to ~$19.5–$20.5B post-guide.
Guidance style: Investment-mode-explicit. The Q1 guide is the cleanest version of "we are investing for FY27/28 returns" management has provided.
Analyst Q&A Highlights
ROIC and FCF Guardrails
- Mark Mahaney, Evercore: Asked for FCF floor commitments and ROIC framework against the capex curve. Olsavsky framed via backlog conversion and AWS unit economics. Jassy reiterated AWS investment math: capacity is monetizing as installed; AI inference will be majority of long-term workloads; custom silicon improves both customer price/performance and Amazon margins. Neither committed to a specific FCF floor.
Assessment: The non-commitment is itself the guidance. Multi-year FCF compression is the explicit trade-off.
Project Rainier and Trainium Roadmap
- Doug Anmuth, JPMorgan: Asked about Project Rainier scaling (500K to 1M chips) and FCF guardrails. Jassy: 500K Trainium2 deployed for Rainier with Anthropic, increasing further; Anthropic also using Trainium2 for other workloads beyond Rainier. Trainium3 just shipping at +40% better price/performance; nearly all supply committed by mid-2026. Trainium4 in build for 2027; "very substantial interest already." Trainium5 in conversations.
Assessment: The chip roadmap visibility through 2027 is the strongest forward guidance. Implies AWS AI revenue conversion well into 2027/28.
OpenAI Relationship
- Ross Sandler, Barclays: Asked about the AI market shape and how the OpenAI relationship might extend. Jassy described demand as "barbelled" — AI labs at one end (gobs of compute), enterprises in productivity / cost-avoidance use cases at the other, and the "middle barbell" (enterprise production workloads, brand-new businesses) being where the largest and most durable demand will eventually settle.
Assessment: The OpenAI mention in Jassy's prepared remarks is the most consequential customer announcement. Doesn't displace the Anthropic relationship; expands AWS's frontier-lab footprint.
Q1 Guide Composition
- Implied questions on Q1 guide softness — Olsavsky preempted in opening remarks: $1B YoY NA cost increase from Amazon LEO satellite launches, plus continued International investment in delivery and sharp-pricing markets.
Assessment: Investment-driven, not demand-driven softness. The market may not parse the difference in the near term.
What They're NOT Saying
- Full-year 2026 op-income outlook: Not provided (consistent with Amazon's quarterly-only guidance practice). But the $200B capex + $1B/quarter LEO cost would imply a flat-to-modestly-down operating income year if AWS revenue tracked the low end of any plausible scenario. This omission is conspicuous and amplifies the multiple compression.
- 2026 FCF range: Not provided. Mark Mahaney explicitly asked. Olsavsky declined.
- OpenAI deal size and structure: Mentioned in customer list. No commercial structure disclosed.
- Trainium revenue contribution as a percentage of AWS: Combined Graviton+Trainium $10B+ disclosed; Trainium-specific share of AWS not broken out.
- 2026 AWS growth target: Not committed. Backlog implies +22-25% achievable.
- Anthropic equity-stake updated valuation framework: No incremental disclosure.
- FTC settlement structural remedies: Settlement amount disclosed in Q3; structural marketplace changes still not detailed.
- Memory chip supply commentary: Newly added to risk language but not addressed substantively in commentary. Possible 2026 cost pressure if HBM tightens.
Market Reaction
- After-hours move (Feb 5): AMZN initially traded modestly higher on the AWS beat, then sold off ~7-11% as the $200B 2026 capex disclosure landed in the press release commentary and on the call. Closed after-hours session approximately −7% to −11% depending on session timing.
- Next-day (Feb 6) early indications: Stock down ~5% in pre-market. Estimated next-day close: −4% to −8%. Multiple desks moving up FY26 revenue / operating income estimates, but the multiple compresses as the FCF model gets reset.
- Volume: Highest single-session AMZN volume in 12 months.
- Analyst reactions (within first 12 hours):
- Sell-side responses split: most desks maintained ratings but moved PTs in a wider band (some up on AWS reaccel + custom silicon; some down on FCF compression)
- Bull camp: AWS beat + reacceleration confirms supply thesis; backlog +40% YoY visibility through FY26-27; Trainium2+3 mathematically justifies capex; NA margin record
- Bear camp: $200B capex unsustainable; Q1 guide soft; FCF compression to $5–15B trough in FY26; multiple compresses meaningfully on visible 2-year FCF compression
The selloff is the textbook "operationally strong, capex-spook" reaction. The print confirmed the AWS reacceleration thesis we have been tracking — but the magnitude of the 2026 capex commitment surprised even the most bullish models. The 8–11% move compresses the multiple on a print where forward earnings power expanded materially. That is the textbook setup for an Outperform call we are willing to make against the post-print weakness.
Street Perspective
Debate: $200B Capex — Strategic Genius or Excess?
Bull view: Pre-commitment of $200B implies management has visibility into AWS demand well into 2027–2028. Backlog +40% YoY corroborates. The AWS-LTV-per-capex math (each $1B converts to ~$300M annualized revenue at ~35% margins) supports incremental ROIC at or above the historical AWS pattern. Custom silicon improves the unit economics further. This is the AWS 2.0 buildout, repeating the original recipe.
Bear view: $200B in a single year exceeds anything any company has ever done. Even rational unit economics break if demand softens by 10–15% in any quarter. The 2-year FCF compression to $5–15B range will compress the multiple by 15–25% before AWS revenue catches up. Mark-to-market risk is real.
Our take: The bull case has the operational data on its side (24% growth, $244B backlog, 1.4M Trainium2 chips landed, OpenAI customer add). The bear case has the multiple-compression timing on its side. Net: this is a 12–18 month patience trade. Outperform with eyes open to multiple risk.
Debate: Trainium / Custom Silicon as the AWS AI Moat
Bull view: $10B+ combined run rate growing triple-digit. Trainium3 supply already 50%+ committed by mid-2026. Trainium4 design wins in progress. Bedrock token usage majority on Trainium = AWS captures both compute and silicon margin. Closer to a 3–5 year structural advantage than a quarter-by-quarter battle.
Bear view: NVIDIA roadmap (Blackwell, Rubin) keeps closing performance gaps; Microsoft Maia and Google TPU v6 are scaling. Custom silicon advantages are inherently transient.
Our take: Bull view supported by data. The "majority of Bedrock token usage on Trainium" disclosure plus the customer-count breadth (100K+ companies) is hard to undo in 18–24 months. Moat thesis is intact.
Debate: Retail Margin — Sustainable at 9% Q4?
Bull view: NA Q4 9.0% margin is structural — driven by network productivity, advertising operating leverage, and 3P unit-mix. Same-day perishables and Amazon Now layered on top. Path to 10%+ in 2026.
Bear view: Q4 is seasonally the strongest margin quarter; full-year 2025 NA margin was 6.9% — Q1/Q2 will likely give back. Tariff pass-through risk remains.
Our take: Full-year NA margin will normalize to 7.5–8.0% in FY26, still a structural improvement from FY25's 6.9%. Bull case wins on trajectory.
Model Update Needed
| Item | Pre-print model | Suggested change | Reason |
|---|---|---|---|
| FY26 revenue growth | +11% | +12-13% | AWS reaccel + retail momentum |
| FY26 op income (ex-charges) | $95–105B | $98–110B | AWS margin durability + ad accel |
| AWS FY26 growth | +19-20% | +21-23% | Q4 +24% + $244B backlog |
| AWS FY26 op margin | 34% | 34-36% | Custom silicon mix benefit |
| FY26 capex | $145B | $200B (guided) | Formal disclosure |
| FY26 FCF | $30-50B | $5-25B | Capex step-up dominates |
| FY27 FCF (recovery) | $50–70B | $60–90B | AWS revenue conversion of FY26 capex |
| Advertising FY26 | +20% | +22-24% | 4 quarters of accel + CTV moat |
| NA op margin (FY26) | 7.0% | 7.5-8.0% | Q4 9.0% record sustaining |
Valuation impact: Forward earnings-power upgrade offsets near-term FCF compression. Net target moves up; the 8-11% selloff creates an attractive entry point for 12-month holders. Patient capital wins; mark-to-market traders may struggle through Q1/Q2.
Thesis Scorecard Post-Earnings
| Thesis Point | Status | Notes |
|---|---|---|
| Bull #1: AWS reaccelerates on AI demand | Confirmed (twice) | +20.2% Q3 → +24% Q4; backlog +40% YoY |
| Bull #2: Retail margin expansion compounds | Confirmed | NA 9.0% record Q4 margin |
| Bull #3: Advertising compounds at 18%+ | Confirmed | Fourth accel quarter; Prime Video 315M |
| Bull #4: Trainium/Anthropic anchor durable AWS AI moat | Strengthened | $10B+ silicon run rate; Trainium3 launched |
| Bull #5 [new]: Same-day perishables + Amazon Now structural growth | Confirmed | 2,300+ cities; India 3× frequency |
| Bear #1: AWS losing relative AI share | Retired | +24% on $142B run rate; OpenAI add |
| Bear #2: Capex compresses FCF | Confirmed (severely) | $200B 2026 capex; FY26 FCF compresses further |
| Bear #3: Tariffs compress NA margin/demand | Retired (for now) | 4 quarters of no signal |
| Bear #4: FTC antitrust overhang | Resolved | $2.5B settlement final |
| Bear #5 [new]: 2026 multiple compression risk | Active | $200B capex resets FCF curve |
Overall: Operational thesis decisively strengthened; valuation thesis weakened by FCF compression visibility. Forward earnings power expanded; near-term multiple risk increased.
Action: Maintain Outperform. The AWS reacceleration sustaining + Trainium silicon moat + advertising compounding + retail margin record together create a multi-year earnings curve that the post-print selloff does not capture. Patient holders are rewarded; the 8-11% gap creates entry. Outperform maintained through FY26.