AWS appears in Signwl's briefs in three distinct postures: as the dominant B200 supplier whose pricing dispersion drives multi-stock analysis; as a custom-silicon producer whose Trainium/Inferentia spot prices are now a leading indicator for the broader Nvidia substitution arc; and as the operator behind the flat-$8.69 H100 on-demand ceiling in us-virginia that anchors much of the H100 OD price tier across global regions.
The B200 supplier. AWS is currently the only public-market B200 quoter at scale. Ohio at $6.05/hr (+136% 30d), Oregon $4.85/hr (+84% 30d), us-east-1 at $2.76/hr (–22% 30d) — three regions, same hardware, 119% range. The internal capacity allocation choices behind these prices ripple into the broader Nvidia-vs-installed-base economics. AWS's capacity decisions are the principal mechanism by which B200 spot prices translate into H100 depreciation outcomes.
The Neuron substitution signal. Trainium spot in us-ohio printed +332% over seven days; Inferentia2 in Stockholm +153% in a single 24-hour window; Trainium2 on-demand rose 13.1% on the week with no corresponding spot move. Per Andy Jassy's commentary: a $225bn Trainium backlog. The interpretation we have arrived at: Neuron-family spot spikes are real-time demand signals from workloads that have already migrated off Nvidia, not noise. AWS is the only hyperscaler where internal demand for custom silicon is visible in spot pricing — Azure Maia and Google TPU show through earnings disclosures rather than spot prints.
The H100 OD anchor. us-virginia H100 on-demand held a perfectly flat $8.69/hr ceiling for four consecutive months — almost certainly a reserved-contract anchor passed through as catalog pricing. The stability is unusual; AWS's pricing posture here defines the "anchor" tier in the H100 fragmentation theme. When this number moves materially, it signals either AWS reserved-contract terms shifting or a broader demand-supply rebalance.
The Bedrock / Olympus context. AWS's internal AI products consume meaningful Neuron + GPU capacity. The visible spot dispersion is partly driven by these internal flows. Future earnings disclosures should clarify whether the Q1 2026 Trainium backlog growth is being driven by external AWS customers or by internal product capacity.
The composite picture: a name with extraordinary cross-positioning — Nvidia's largest customer, simultaneously its most credible substitute supplier, simultaneously the cloud operator setting the marginal H100 reserved-contract price. Each posture interacts; understanding the AWS read requires holding all three together.