Amazon (AMZN)
Hyperscaler (AWS) + e-commerce + the Anthropic partnership + Trainium chips — the largest cloud + a custom-silicon AI build
AI exposure 8% — derived from reported segment revenue, not asserted.
Where it sits
AI Platforms → Hyperscalers & Cloud Platforms · figures rebased to Q2 2026, reported 2026-07-30
No business is a majority here. AWS — AI cloud (Trainium/Inferentia, Bedrock) is 17% of revenue — placement rule 2 puts the primary with the largest business, so Hyperscalers & Cloud Platforms. The remaining 67% (E-commerce / retail / marketplace) is deliberately off-map.
Reported segments
- AWS (cloud) — 17% of revenue
- Online stores + Physical + 3P seller services (e-commerce) — 67% of revenue
- Advertising services — 9% of revenue
- Subscription services (Prime) — 7% of revenue
Primary B6 rests on AWS (17%, the AI cloud + most of profit), NOT the 67% e-commerce (which is off-map by kind). Classic dominant-AI-relevant-revenue (not largest-segment) case.
How the 8% is built
- AWS (cloud) — 17% of revenue × 30% AI-driven (estimate). The disclosed >$25B AI run rate is ~15% of AWS's $169B run rate, but that figure counts only named AI services (Bedrock/SageMaker/AI) and excludes the accelerated-compute EC2 capacity AI workloads consume — which is what drove AWS's acceleration to +37% and the RPO jump to $496B. 30% is held as the estimate spanning…
- Online stores + Physical + 3P seller services (e-commerce) — 67% of revenue × 3% AI-driven (estimate). AI in logistics forecasting and recommendations is long-standing internal ML, not an AI revenue line
- Advertising services — 9% of revenue × 8% AI-driven (estimate). Ads grew 26% to $19.8B with AI targeting and generative creative cited as contributors, but Amazon does not size the AI-attributable portion
- Subscription services (Prime) — 7% of revenue × 2% AI-driven (estimate). Prime value rests on shipping and video; minimal direct AI linkage
Investment read
Moat. AWS hyperscale leadership (the #1 cloud) + e-commerce/logistics scale + the Anthropic partnership + Trainium/Inferentia custom silicon; the cloud-infra moat.
Bottleneck / pricing power. Strong pricing power (AWS scale + switching costs + the custom-silicon cost lever); AWS is the profit engine funding the AI build.
Role in the AI stack. The largest hyperscaler-cloud + custom-AI-silicon layer (AWS + Trainium).
Connected companies
- Talen Energy (TLN) — AMZN is supplied by: 1,920 MW Susquehanna nuclear PPA with AWS through 2042 (front-of-the-meter since the Apr 2026 transition; up to ~$1.4B/yr at full volume)
- NVIDIA (NVDA) — AMZN is supplied by: GPUs for AWS
- Marvell Technology (MRVL) — AMZN is supplied by: 5-yr multi-generational AWS agreement: custom AI silicon (Trainium) + optical DSPs/networking
- Ciena (CIEN) — AMZN is supplied by: optical transport for cloud/AI expansion (pair not publicly named; AWS exposure partly via carrier networks built on Ciena)
- Snowflake (SNOW) — AMZN supplies: AWS compute/storage underpins Snowflake's data cloud
- Advanced Micro Devices (AMD) — AMZN is supplied by: EPYC CPUs powering AWS instances
- Intel (INTC) — AMZN is supplied by: custom Xeon Scalable CPUs across EC2 (C7i/M7i/R7i)
- Equinix (EQIX) — AMZN is supplied by: colocation hosting AWS Direct Connect on-ramps in 45+ markets
- Digital Realty Trust (DLR) — AMZN is supplied by: data-center space hosting AWS Direct Connect on-ramps (100G nodes in DLR facilities)
- Seagate Technology (STX) — AMZN is supplied by: mass-capacity HDDs for AWS cloud storage
- Western Digital (WDC) — AMZN is supplied by: HDDs for hyperscale cloud (AWS)
- NetApp (NTAP) — AMZN is supplied by: FSx for NetApp ONTAP first-party AWS service
- Everpure (fmr Pure Storage) (P) — AMZN is supplied by: AWS partnership (Cloud Block Store on AWS marketplace); no evidence AWS deploys Pure arrays internally
- Broadcom (AVGO) — AMZN is supplied by: merchant switch silicon (Tomahawk) for AWS self-built network gear; Trainium ASIC work is Marvell/Annapurna, not Broadcom
Others in Hyperscalers & Cloud Platforms
All 174 companies · What moved this quarter · How the map is built