Deep Research

Amazon.com (AMZN)

4 min read

Snapshot

  • Ticker: AMZN (Nasdaq)
  • Bucket: Compute & AI Megacaps
  • Q1 2026 position: 953,560 shares, $198.6M, 3.97% of 13F — roughly the #6 equity position (#7 line overall behind the QQQ put)
  • HQ: Seattle, Washington
  • What it does: Global e-commerce and logistics platform plus AWS, the largest cloud-infrastructure provider — and, increasingly, a vertically integrated AI-silicon vendor via Trainium.

Business Overview

Amazon runs three reporting segments: North America and International retail (first-party commerce, third-party marketplace services, advertising, Prime subscriptions) and AWS. FY2024 revenue was 575B in 2023, with AWS contributing 45.6B — an operating margin around the low-40s and the large majority of total company operating profit — while management laid out a ~$200B capex plan for 2026. Advertising remains the fastest-compounding high-margin line inside retail.

The Q1 2026 print is what the bulls had been waiting for: revenue of 150B annualized run rate. The cost is enormous: Q1 capex hit 1.2B, and reported net income was inflated by a $16.8B unrealized gain on Amazon’s Anthropic stake. Management guided to significant further investment through 2026.

The strategic story inside AWS is custom silicon. Per the Q1 2026 call, Trainium2 is largely sold out, Trainium3 (ramping H2 2026) is nearly fully subscribed, and much of Trainium4 is already reserved; AWS cited >20B annualized business growing at triple-digit rates. Management frames Trainium as saving tens of billions of capex annually and worth several hundred basis points of operating-margin advantage versus buying merchant inference silicon.

Financial Trajectory

PeriodRevenueYoYNotes
FY2023$574.8B+12%—
FY2024$638.0B+11%AWS $107.6B (+19%)
FY2025n/a (not sourced)—AWS op. income ~200B FY26 capex plan
Q1 2026$181.5B+17%AWS +28% (~44.2B; FCF ~$1.2B (−95%)

Why Atreides Owns It

Amazon is the megacap expression of Baker’s “watts and wafers” framework: it owns the watts (one of the world’s largest datacenter and power-procurement footprints), and with Trainium it increasingly controls its own wafers. At Sohn New York 2026 Baker called Trainium “the most undervalued AI chip” — not Nvidia, not Google’s TPU. His technical argument: only two operational switched scale-up networks exist in the world, one moving data between Nvidia GPUs and one between Trainiums, a capability he says TPUs lack (“Google invented the MLPerf benchmark but doesn’t submit TPU results to its own benchmark”). He predicted Trainium’s position in 2026 will be what TPU’s was in 2025 once Trainium3 ramps in the second half.

That makes the AMZN line partly an AI-silicon position wearing a retailer’s clothes: the market caps Amazon on retail margins and AWS growth, while Baker sees an underpriced vertically integrated accelerator franchise plus the Anthropic stake (the $16.8B Q1 mark-up validates the carry). The position’s stability — held every quarter at 660–950K shares while Atreides exited Microsoft twice and dumped Meta and AMD — signals it is the fund’s preferred hyperscaler. The 39% share-count add in Q1 2026 came in the same quarter Baker was making the Trainium case publicly.

Position History

QuarterTypeShares/NotionalValue% of 13F
Q4 2024Common680,737$149,346,8903.27%
Q1 2025Common689,231$131,133,0903.98%
Q2 2025Common718,465$157,624,0364.37%
Q3 2025Common658,886$144,671,5992.82%
Q4 2025Common687,799$158,757,7651.94%
Q1 2026Common953,560$198,597,9413.97%

This is one of the steadiest lines in the book — held continuously across all six quarters with only modest trims, in stark contrast to the fund’s signature whipsaws (Astera, Micron, NVDA). The Q1 2026 add of ~266K shares (+39%) is the largest single-quarter change in the series and coincides with Baker’s public Trainium advocacy. No options overlay has ever appeared on the name.

Risks

  • Capex digestion: ~$200B of planned 2026 capex against collapsing free cash flow; if AI demand decelerates, Amazon owns the depreciation.
  • Trainium thesis is early: Trainium’s traction is heavily Anthropic-anchored; if frontier labs consolidate on Nvidia/TPU or Anthropic diversifies, the “most undervalued chip” case weakens.
  • Accounting optics: Q1 2026 net income was flattered by a $16.8B unrealized Anthropic gain — underlying earnings quality is noisier than the headline.
  • Retail cyclicality and tariffs: the majority of revenue is still low-margin commerce exposed to consumer weakness and trade policy.
  • Regulatory: ongoing FTC antitrust litigation and EU platform regulation target the marketplace/Prime/logistics bundle.
  • Crowding: AMZN is a consensus megacap long; Atreides’ edge here is thesis differentiation (Trainium), not discovery.

Sources

Backlinks

Graph View