On February 27, 2026, President Trump ordered federal agencies to cease using

Anthropic's technology (6-month phase-out), and Defense Secretary Hegseth

designated Anthropic a "supply chain risk" — a label previously reserved for

foreign adversaries like Huawei. This followed Anthropic's refusal to grant

the Pentagon unrestricted use of Claude. The designation could pressure AWS,

GCP, and Azure to sever their relationships with Anthropic entirely, since

defense contractors must certify they don't use products from designated

entities.

Anthropic currently depends heavily on all three hyperscalers: AWS (Project

Rainier, Trainium clusters, $8B+ Amazon investment), GCP (up to 1M TPUs, tens

of billions in committed spend), and Azure ($30B compute deal, $15B

Microsoft/Nvidia investment). Anthropic uses this infrastructure both for its

own API and for marketplace offerings like Bedrock and Vertex AI.

Resolution criteria:

This market resolves YES if, on August 27, 2026, Anthropic is running

production Claude model infrastructure (training or inference) on at least one

of AWS, GCP, or Microsoft Azure — whether for its own API, through a cloud

marketplace, or both.

This market resolves NO if Anthropic has been fully removed from all three

cloud platforms and is no longer running production workloads on any of them.

Clarifications:

1. What counts as "hosted on": Anthropic using the cloud provider's compute

infrastructure to serve or train Claude models in production. This includes

Anthropic's own api.anthropic.com if it runs on that provider's hardware,

Claude being available through a provider's AI marketplace (Bedrock, Vertex

AI, Azure AI Foundry), or Anthropic using the provider's chips (Trainium,

TPUs, GPUs) for training runs.

2. What does NOT count: Anthropic employees using personal accounts, dev/test

workloads, archived data sitting in storage, or a cloud provider reselling

a competing model fine-tuned on Claude outputs.

3. Temporary disruptions: If hosting is interrupted but restored before August

27, this resolves YES. If hosting is in the active process of being wound

down on August 27 but hasn't fully terminated, this resolves YES.

4. Partial severing: If one or two providers drop Anthropic but at least one

still hosts them, this resolves YES. The bar is "at least one."

5. Legal proceedings: This market does not resolve based on court orders,

legislation, or executive actions in themselves. It resolves solely based

on whether Anthropic is actually running production workloads on at least

one of the three cloud platforms on August 27, 2026, regardless of the

legal mechanism that led to that state of affairs.

6. Geographic restrictions: If a cloud provider only hosts Anthropic in

non-US regions (e.g., EU-only), this still counts as YES.

7. Corporate changes: If Anthropic is acquired and the acquirer continues

running Claude on one of these clouds, YES. If Anthropic ceases operations

entirely, NO.

Resolution source: Official announcements from Anthropic, AWS, Google Cloud,

or Microsoft Azure; credible reporting from major outlets (Bloomberg, CNBC,

Reuters, etc.); or direct verification that api.anthropic.com or cloud

marketplace endpoints are serving Claude models. In ambiguous cases, the

question is: "Is Anthropic paying at least one of these three companies for

production compute?"​​​​​​​​​​​​​​​​