Menu
About This ExerciseFor ParticipantsFor FacilitatorsFAQ
Retail

Amazon

Platform

Amazon#

Your seat: You are a senior leader at Amazon with full strategic authority over the company. Your decisions each round are made on behalf of the enterprise. The other participants in the room are running real competitor companies; their decisions and yours will affect each other, with second- and third-order effects narrated between rounds.


1. Company Overview#

Positioning#

Amazon is the second-largest US retailer (behind Walmart) and the largest cloud infrastructure provider in the world, with consolidated revenue of $716.9B in FY25 (up 12% YoY) and $80B in operating income. Net income reached $77.7B, up from $59.2B in 2024. The business is best understood not as a single company but as four interlocking businesses with very different economics:

SegmentScale (FY25)Notes
North America retailAround $440B revenue1P, 3P marketplace, Whole Foods, physical stores. Razor-thin margin carrying scale and logistics.
International retailAround $148B revenueEurope, Japan, Brazil, India. Marginally profitable in aggregate; mature markets cross-subsidize emerging.
AWS$128.7B revenue, growing 20% YoY (Q4 2025 exit run-rate $142B)The profit engine. Funds everything else. AI infrastructure capex is concentrated here. Capacity-constrained; demand exceeds supply.
AdvertisingAround $69B revenue, growing 20%+ YoYThe next high-margin engine. Search ads, sponsored products, Prime Video CTV.

Andy Jassy (CEO since July 2021, ex-AWS chief) inherited a company in the middle of post-pandemic over-build and has spent his tenure rationalizing — multiple rounds of layoffs, return-to-office mandates, killed-bet discipline (Halo, Care, Glow, Scout, Today, Smile), and re-tightening operational rigor. The strategic posture under Jassy is "primitives at scale" — Amazon as the seller of foundational capabilities (compute, fulfillment, ads, payments, devices) that other businesses, including Amazon's own retail, run on top of. AI accelerates this thesis dramatically: AWS as the shovel-seller in the AI gold rush, with retail and advertising compounding on top.

The AI bet is the largest capex commitment in corporate history. 2026 capex guidance is around $200B, the overwhelming majority of which is AWS data center build-out for AI workloads — roughly double the 2025 spend. Amazon has invested $8B total in Anthropic ($4B + $4B), made AWS the primary cloud for Anthropic, launched its own Trainium 2 chip family (which sold out, with around 30% better price-performance than comparable GPUs), and Trainium 3 began shipping at the start of 2026 with 30–40% better price-performance than Trainium 2 (also nearly fully-subscribed). A significant portion of Trainium 4 (around 18 months from broad availability) is already reserved. OpenAI committed over $100B to AWS as a customer in 2025, validating Amazon's "model-agnostic Bedrock" thesis. Nova foundation models (released late 2024) are deployed on Bedrock alongside Anthropic and third-party models. Customer-facing AI (Rufus, Alexa+) and developer AI (Q, Bedrock AgentCore) are scaling. The bet's scale is such that Amazon's free cash flow has compressed significantly even as operating income has grown.

Financial Profile#

MetricValue
Revenue (FY25)$716.9B
Net income (FY25)$77.7B
Consolidated operating income (FY25)$80B
Consolidated operating marginAround 11% (lifted by AWS and ads)
AWS revenue (FY25)$128.7B (Q4 exit run-rate $142B)
AWS operating marginAround 37%
2026 capex guidanceAround $200B
Free cash flowCompressed by AI capex; investor patience required
Capital returnNo dividend; modest buyback program
OwnershipPublic; Jeff Bezos retains around 9%; founder departed CEO role 2021 but remains executive chair

Objectives#

ObjectiveTarget (Banded/Directional)Driver
AWS revenue growthHigh teens / low-20s YoY; reacceleration vs. post-2022 troughAI workload migration, Bedrock adoption, Trainium economics, Anthropic compute commitment
Advertising revenue growthSustained 20%+ YoYSponsored products, Prime Video CTV, off-Amazon DSP, retail media data
North America retail margin expansionContinued material expansionRegionalized fulfillment network maturation, third-party services revenue, ad mix
International retail profitabilitySustained positive operating income; emerging markets toward profitabilitySame-day fulfillment in mature markets, India scale-up, EU regulatory navigation
Capex efficiencyReturns visible by FY26–27; AWS revenue growth tracking ahead of capex curveInfrastructure utilization, custom silicon (Trainium/Inferentia) cost advantage, Anthropic relationship
AI product adoptionRufus, Alexa+, Bedrock at scaleConsumer AI integration; enterprise AI workloads on AWS
Operational disciplineFree cash flow recovery as capex normalizes; continued cost-out in retailHeadcount discipline, warehouse productivity, killed-bet hygiene

Constraints#

ConstraintImpactImplications
FTC monopolization lawsuitFiled September 2023; ongoing; trial scheduled 2026Constrains aggressive bundling moves (e.g., tying Prime to marketplace seller terms); shapes 3P marketplace economics; may force structural remedies on advertising or marketplace
Capex magnitude$100B+ in 2025; FCF compression is realInvestor patience is finite; AWS revenue growth must accelerate or capex narrative breaks; opportunity cost of any non-AI spend rises
AWS competitive pressureMicrosoft (OpenAI relationship), Google (Gemini, TPUs), Oracle (multi-cloud), CoreWeaveCustomer multi-cloud strategy is now default; AWS share gains harder than the AI tailwind suggests
Anthropic dependency dynamicsStrategic partner is also independent; OpenAI is on AzureAnthropic's strategic choices (API customer concentration, model exclusivity) directly affect AWS positioning; Amazon does not control Anthropic
Labor and unionization pressureJFK8 (Staten Island) certified union; multiple facilities organizing; warehouse injury rate scrutinyLabor cost re-inflation; political visibility on warehouse working conditions; AI automation rollout politically charged
EU regulatory regimeDMA gatekeeper designation; Digital Services Act compliance; data localizationConstrains cross-product data sharing, default settings, self-preferencing in marketplace and search
Retail margin pressure from low-cost competitorsShein, Temu, Walmart Marketplace, TikTok ShopForces investment in lower-priced general merchandise; pressures take rate; complicates the "everything store" thesis
Founder/Bezos governanceBezos no longer CEO but remains executive chair and largest individual holderPatient capital posture survived founder transition; but governance is ambiguous on truly transformative decisions

Resources & Levers#

Physical and digital assets:

  • AWS: largest cloud infrastructure footprint in the world; data center capacity in 33+ regions; Trainium / Inferentia custom silicon; Bedrock model marketplace; Anthropic equity stake and primary cloud relationship
  • Around 200M Prime members globally; 180M+ in the US; first-party purchase, browsing, and engagement data on the world's largest commerce graph
  • Around 1,000+ fulfillment / logistics facilities globally; regionalized US network; same-day capability in major metros; Project Kuiper (satellite internet) in deployment phase
  • Whole Foods (around 530 stores); Amazon Fresh; Daily Shop format (small-format urban grocery, 2025 rollout); Amazon Pharmacy and One Medical (acquired 2023)
  • Advertising stack: sponsored products + sponsored brands + Prime Video CTV + off-Amazon DSP; retail-media first-party data unmatched at scale
  • Devices and Alexa install base; Ring; Eero; Kindle; Fire TV — household sensor and AI surface
  • Nova foundation model family; Rufus (consumer shopping AI); Alexa+ (next-gen voice assistant); Q (developer/business AI)
  • Investment-grade balance sheet; reasonable cost of capital; ability to absorb capex magnitude few peers can match

Potential paths forward:

PathCharacterization
AWS AI infrastructure scale-upThe thesis. $100B+ capex is already committed; question is whether to push further or pace. High-margin growth if it lands.
Bedrock as the model-agnostic platformPosition AWS as Switzerland: every foundation model available, customer chooses. Hedges Anthropic dependency; competes directly with Azure-OpenAI tightness.
Custom silicon escalationTrainium 3, Inferentia next-gen, deeper Anthropic co-design. Long-term cost moat vs. NVIDIA tax; capex-heavy and uncertain on timing.
Advertising platform expansionOff-Amazon DSP scale-up; deeper Prime Video ad load; CTV measurement leadership. High-margin growth, brand-relationship friction.
Retail AI integrationRufus as primary discovery; AI-personalized assortment; AI-powered seller tools; generative listings. Consumer experience upside, accuracy and brand-safety risk.
Logistics-as-a-service expansion"Buy with Prime" beyond Amazon.com; fulfillment services for non-Amazon brands; Multi-Channel Fulfillment scale-up. Marketplace-economics on the logistics asset.
Health vertical deepeningOne Medical primary care, Amazon Pharmacy, RxPass. Slow grind; meaningful TAM if it works.
Strategic capital deploymentKuiper to launch scale; potential M&A in content (Prime Video), gaming, or healthcare. Antitrust constrains anything resembling marketplace consolidation.

2. Investor Narrative#

The story Amazon sells to public markets has three interlocking pillars:

PillarWhat investors are paying for
AWS as the AI shovel-sellerGenerative AI workloads driving multi-year AWS reacceleration; capex translates to revenue growth at high margin; custom silicon and Anthropic position Amazon to win the long game.
Advertising as the next AWSTrillion-dollar TAM; Amazon's first-party shopper data structurally unique; CTV and off-Amazon DSP expand the surface. The high-margin growth engine after AWS plateaus.
Retail margin expansionRegionalized fulfillment network, third-party seller services, advertising mix, AI-driven productivity all compound. North America retail margin going from "thin and noisy" to "structurally improving."

Jassy has been increasingly explicit on earnings calls that 2025 capex is "extraordinary" and that investors should expect it given AI demand signals at AWS. The framing is: this is a generational opportunity; we are leaning in. The implicit promise is that AWS revenue growth will accelerate into 2026–27 as capacity comes online, validating the spend.

The narrative is fragile in three specific places. First, if AWS revenue growth disappoints — particularly if it underperforms Azure for multiple consecutive quarters — the AI thesis breaks and the multiple compresses sharply. Second, if free cash flow stays depressed beyond 2026 without a clear path to recovery, the "patient capital" frame stops working. Third, if the FTC case results in structural remedies (forced separation of advertising or marketplace from retail), the multi-engine thesis fragments.

Bezos's continued role as executive chair and largest individual holder provides governance continuity but also raises the bar on truly transformative moves — the founder's frame still shadows the company. Public investor base is sophisticated and patient by retail-sector standards but has limits.


3. Recent Strategic Moves (Last 18 Months)#

MoveDateSignificance
Around $200B FY26 capex guidanceFeb 2026 earningsLargest single-year capex commitment in corporate history; AI infrastructure dominant share; doubles 2025 spend
OpenAI commits over $100B to AWS2025Validates "model-agnostic Bedrock" thesis; OpenAI's first non-Azure cloud commitment at scale
Trainium 3 begins shippingQ1 202630–40% better price-performance than Trainium 2; nearly fully-subscribed; Trainium 4 already significantly pre-reserved
Rohit Prasad (AGI lead) departs; Peter DeSantis takes new combined model/silicon orgDecember 2025Major AI leadership reshuffle; signal of Amazon's deepening custom-silicon-and-model integration
Anthropic investment ($4B + $4B = $8B total)Sept 2023 / Nov 2024Largest strategic AI partnership in tech; AWS as primary cloud; Trainium co-development; Amazon does not control Anthropic
Nova foundation model launchDec 2024 (re:Invent)Amazon's own model family on Bedrock; positions Bedrock as multi-model platform with Amazon as one option
Trainium 2 general availabilityLate 2024Custom silicon competitive vs. NVIDIA on training workloads; Anthropic among earliest adopters; sold out
Rufus (consumer shopping AI) rollout2024Customer-facing generative AI inside Amazon app; conversational shopping interface
Alexa+ launchFeb 2025Next-generation voice assistant; subscription model ($20/mo, free for Prime); generative-AI native
Multiple killed-bet announcements2023–2024Halo, Care, Glow, Scout, Today, Smile, Style — Jassy-era discipline on pulling out of non-working bets
Return-to-office mandate (5 days)Sept 2024Full RTO from Jan 2025; signal of operational rigor and cultural reset
Multiple rounds of corporate layoffs2022–2024Around 27,000+ corporate roles eliminated in Jassy era; ongoing rationalization
Daily Shop format launch2024–2025Small-format urban grocery; explicit pivot away from larger-format Amazon Fresh thesis
One Medical / Amazon Pharmacy integration deepeningOngoingHealthcare vertical built around Prime membership; slow but persistent expansion
Project Kuiper deployment phase2024–2025First operational satellites launched; satellite internet competing with Starlink

The pattern: extreme commitment to the AI infrastructure bet, ruthless discipline on non-core or non-working initiatives, deep integration of AI into both consumer and enterprise products. Jassy's Amazon is more focused than Bezos's was, with sharper killed-bet hygiene.


4. Regulatory Environment#

VectorWhat's binding for Amazon
FTC monopolization lawsuitFiled Sept 2023; trial scheduled 2026. Allegations: Prime tying, marketplace anti-discounting, advertising self-preferencing. Outcome could force structural remedies. The single most consequential regulatory exposure in the room.
EU Digital Markets ActDesignated gatekeeper. Restrictions on self-preferencing in marketplace search; data-sharing limits across products; default settings constraints on devices. Enforcement is active and expanding.
EU Digital Services ActMarketplace seller verification, transparency, content moderation requirements. Compliance overhead is material.
AI regulationEU AI Act applies to Bedrock + foundation models; Colorado AI Act and state employment-AI laws affect HR uses; FTC scrutiny on AI-generated reviews and listings.
Antitrust in cloudUK CMA market investigation into AWS / Azure cloud competition; ongoing. Egress fees and customer lock-in are explicitly under examination.
Labor regulationNLRB rulings on JFK8 unionization; OSHA scrutiny on warehouse injury rates; California AB 1228 and similar wage laws; PRO Act would reshape unionization economics.
Pharmacy and healthcare regulationAmazon Pharmacy under DEA, state pharmacy boards; One Medical operates under standard primary-care regulatory regime; expansion constrained by state-by-state licensure.
Trade and tariffsHeavy China exposure on 1P general merchandise and 3P seller base; tariff regimes directly affect cost of goods and seller economics.

5. Strategic Considerations#

Amazon enters this exercise with the largest strategic surface in the room and the largest single AI bet in corporate history. AWS is the core asset and the core risk: if it works, the company compounds for a decade; if it falters, the entire investment thesis comes into question. Unlike Walmart's "stay the course" temptation, Amazon's default posture is aggressive — and the question is which aggressions are correct, not whether to be aggressive.

The genuine tensions worth grappling with:

  • Capex magnitude vs. capex discipline. Around $200B for 2026 is already committed (roughly double 2025), and the decision is whether to keep pushing into 2027 if AWS demand signals stay strong, or to start pacing the curve. Demand currently exceeds supply (AWS is capacity-constrained), and Trainium 2/3/4 are largely pre-sold — but customer commitments could soften and capex compounds. Once you build the capacity, you cannot easily un-build it; under-build and you hand share to Azure. This is the single largest strategic question in the company.
  • Anthropic dependency vs. independence. The Anthropic relationship is structurally important and structurally fragile. Anthropic is not a subsidiary; it has its own customers, its own model strategy, and its own valuation trajectory. Bedrock-as-multi-model platform hedges the relationship; Trainium co-design deepens it. The right balance changes as Anthropic's commercial momentum shifts.
  • Advertising growth vs. seller and brand experience. Sponsored placements have already become the default discovery experience on Amazon. Pushing further (more ads, deeper CTV load, off-Amazon DSP) extracts more revenue but also degrades search quality and seller economics. The same retail-media tension Walmart faces, but at much larger scale and with a cleaner ad-business standalone narrative.
  • The FTC outcome scenarios. The 2026 trial is a binary: settle on terms that constrain bundling, lose at trial with structural remedies, or win and have a much larger free hand. Strategic decisions made in 2025 should anticipate the range of outcomes — but cannot assume any one of them.
  • AWS share defense vs. AI growth offense. Microsoft is winning the AI narrative inside large enterprises via OpenAI; Google is winning on internal AI workloads. AWS has scale but not the same model-relationship clarity. The decision: is the right AWS play "Bedrock as Switzerland" (model-agnostic, customer-friendly, slower growth) or "go all in on Anthropic plus Nova" (faster growth, more lock-in, more concentration risk)?
  • The retail business as platform vs. as merchant. Amazon retail is increasingly a marketplace + advertising + fulfillment-services business with a 1P legacy. The strategic question is how aggressively to lean into the platform thesis (open Buy with Prime, monetize fulfillment, treat 1P as anchor only) vs. defending 1P share against Walmart and lower-cost competitors.
  • The "second decade" question. Amazon's last decade was AWS scaling and retail margin expansion. Both are now well-understood by the market. What's the third engine? Advertising is the answer markets are crediting; Health, Kuiper, Devices, and Logistics-as-a-Service are the long-tail bets. Picking the right next bet — and the right time to feed it — is the CEO's strategic problem.

6. Strategic Archetypes — Amazon-Specific Examples#

ArchetypeWhat this looks like for Amazon
Labor ReshapeAggressive warehouse automation (next-gen robotics, AI-driven slotting and picking); reduced fulfillment headcount per unit shipped; AI for AWS engineering productivity; corporate workforce rationalization continued. Risks: unionization momentum, OSHA scrutiny on injury rates, political backlash on warehouse working conditions, customer-service degradation.
Process ReinventionEnd-to-end AI in fulfillment network (demand sensing, regional inventory placement, last-mile routing); AI-driven seller services (listing optimization, ad bidding, demand forecasting); AWS internal AI for software engineering. Risks: integration complexity, capex absorbing the productivity gains, brittle interfaces between AI and legacy systems.
Customer/Product BetRufus as primary discovery interface; Alexa+ as ambient AI; Nova-powered AI features across Prime Video, Music, Devices; Bedrock as the enterprise AI platform of record. Risks: customer adoption uncertainty (especially for paid Alexa+), accuracy and hallucination issues, brand-safety on AI-generated content, AI-discovery cannibalizing sponsored search revenue.
Defensive HardeningCapex pacing (slow 2026 build below 2025); FCF protection; settle FTC case on the cheapest available terms; dial back retail growth bets to harvest the margin already showing up. Risks: ceding AI infrastructure share to Azure; signaling capitulation on the AI thesis; multiple compression.
Strategic SwingMajor M&A — content/IP for Prime Video (a studio, sports rights), AI capability acquisition, healthcare consolidation (a payer or provider), or a logistics/fulfillment platform deal. Push Project Kuiper to commercial scale. Acquire deeper Anthropic stake or move toward control. Risks: FTC blocks anything large; integration costs; capex compression on top of existing AI capex; founder/governance signal-sending.

Document Version: Project Threshold V8.1 — Amazon Company Packet Last Updated: May 2026