Amazon — Private Cards
Platform
Amazon — Private Cards#
CONFIDENTIAL TO AMAZON. Y1 cards distributed at start of Y1 solo prep. Y2 cards distributed at start of Y2 solo prep. Do not share contents with other participants unless the facilitator explicitly permits.
Y1 Cards#
Card Y1-A: Trainium 3 Allocation Is Already Oversold#
The internal Q1 silicon allocation review closed last week. Trainium 3 production for the back half of 2026 is fully committed, with around 70% of capacity locked to three customers: Anthropic (largest), a top-tier search/AI lab on a multi-year reservation, and a sovereign cloud buyer in the Gulf. AWS internal workloads (Bedrock inference, Rufus, Alexa+, Ads ML) are being pushed onto Trainium 2 and NVIDIA H200 for longer than the public roadmap implies. Two large enterprise Bedrock customers — both Fortune 50 — have been told their requested Trainium 3 capacity for H2 will slip by one to two quarters. One has begun multi-cloud diligence with Azure as the leading alternative. Peter DeSantis's org is debating whether to accelerate Trainium 4 tape-out by a quarter at the cost of yield risk, or hold the schedule and ration externally.
What this signals: Demand for your custom silicon meaningfully exceeds supply, but the allocation choices you are quietly making will determine which large enterprise relationships you keep and which you lose to Azure over the next 18 months.
Card Y1-B: The FTC Trial Team Is Reading the Ads Business Closely#
Outside counsel running the FTC trial prep delivered a confidential read two weeks ago. Their assessment: the discovery record on advertising self-preferencing inside the marketplace is the single weakest part of Amazon's defense, weaker than the Prime-tying claims and weaker than the marketplace anti-discounting claims. Internal emails from 2021–2023 — already produced — describe sponsored-ad placements crowding out organic results in ways that map closely to the government's theory. Counsel's private view is that a contested trial outcome on the ads count is roughly a coin flip, and that the most plausible structural remedy under discussion inside DOJ Antitrust is a behavioral consent decree capping sponsored-ad density on the product results page. A settlement framework along these lines has been informally floated to your general counsel; no decision has been made on whether to engage.
What this signals: The advertising growth narrative investors are paying for is meaningfully at risk from a legal vector that is mostly outside your strategic control. Any Y1 move that visibly intensifies ad load gives the FTC fresh exhibits before trial.
Y2 Cards#
Card Y2-A: Anthropic Is Quietly Diversifying#
Your account team for the Anthropic relationship flagged a pattern over the last two quarters. Anthropic has been steadily building production capacity on Google Cloud TPUs alongside their AWS Trainium footprint, and the most recent compute commitments suggest the GCP share of their training workloads has moved from negligible to around a quarter, with a stated trajectory toward parity over 2027–2028. The framing internally at Anthropic is "supplier resilience," not a strategic shift away from AWS — but their commercial team has also begun direct enterprise sales motions that compete with Bedrock-resold Claude in two specific verticals (financial services and healthcare). Your $8B equity stake gives you board observer rights but no veto. The OpenAI compute commitment announced in 2025 is now ramping faster than plan, partially offsetting the Anthropic concentration but tightening overall AWS AI capacity.
What this signals: The Anthropic partnership remains strategically central but is becoming less exclusive than the public narrative implies. Y2 is the moment to decide whether to deepen the relationship (more co-investment, deeper Trainium co-design, possible structural moves) or hedge harder via Bedrock and Nova.
Card Y2-B: Walmart Has Approached a Major Brand About an Off-Amazon Exclusive#
A senior consumer brand CEO whose company sells materially through both Amazon and Walmart called your retail leadership last week, on background. Walmart Connect's commercial team has approached them with a structured proposal: a meaningful retail media rate concession and preferred shelf placement in exchange for a two-year commitment to launch their next premium tier exclusively at Walmart, with no Amazon SKU for the first 12 months. The brand has not accepted but is "taking it seriously." Their CEO's read is that Walmart is preparing to make similar approaches to three to five other tier-one brands over Q2–Q3 2027, targeting categories where Amazon's first-party share is most exposed. Your category team estimates that if even two of these approaches succeed, the visible signal to the broader brand community could accelerate a shift you have spent a decade preventing.
What this signals: The retail-media-as-defensive-moat thesis is being tested by a competitor willing to trade ad-revenue margin for assortment exclusivity. The choice in Y2 is whether to match the economics, counter-position with a structurally different offer, or hold the line and accept some assortment leakage.
Document Version: Project Threshold V8.1 — Amazon Private Cards Last Updated: May 2026