Cloudflare Workers AI

Serverless inference on Cloudflare GPUs, with a catalog of third-party open-source models and customer-supplied content. Cloudflare’s data-use commitments apply to Workers AI only; AI Gateway may send content to other providers under their own terms.

Score calculation

(8 + 4) ÷ 2 = 6.0

2 of six dimensions rated. Unrated dimensions and BDS status are excluded.

Full scoring method ↗

Setup details

Managed inference API and model catalog for Workers AI. Excludes training of the listed third-party models and AI Gateway requests processed by another provider.

Energy

↑

Reasoning & sources

Cloudflare’s 2025 impact report says its Gen 12 servers deliver 84% more requests per kW than Gen 11 and describes replacing 2,700 older servers with 1,661 newer ones. This is company network efficiency evidence; it does not attribute electricity, emissions, or water to Workers AI, so the service footprint remains unrated. For a deployment, request or export the selected service and region’s usage-linked footprint when available; do not treat company-wide renewable matching as per-request impact.

Workload distribution and infrastructure energy/water attributable to Workers AI remain undisclosed.

Cloudflare Impact Report 2025: Network ↗

Training rights

↑

Reasoning & sources

Cloudflare says it does not train its own LLMs or Workers AI customer content, and that the catalog contains third-party pretrained models subject to their licenses. Corpus rights therefore remain model-specific. Select the model first and review its license plus corpus-provenance disclosures; prompt no-training terms do not clear the model’s original dataset.

The service pages do not establish provenance, consent, or licensing for every upstream model training corpus; review each model’s terms separately.

Privacy

↑

For standard stateless inference, Cloudflare says it retains prompt and output content only as long as needed to process and return the response and does not train on customer content. Stateful storage, user-enabled logging/caching, batch, files/adapters, legal obligations, and third-party routing are exceptions.

Reasoning & sources

The published commitment is narrowly scoped to Workers AI stateless inference and excludes training; feature storage, metadata, legal exceptions, and third-party routing are clearly identified. Check the exact API/web surface, plan, region, retention setting, routing, and stateful features in the applicable terms before sending sensitive content.

The published commitment is narrowly scoped to Workers AI stateless inference and excludes training; feature storage, metadata, legal exceptions, and third-party routing are clearly identified.

The protection depends on feature and configuration. Operational metadata may be retained; AI Gateway requests sent to other providers follow those providers’ practices.

Your Data and Workers AI ↗

Openness

↑

Workers AI offers 50+ third-party open-source models and allows custom model requests, but Cloudflare operates the serverless GPU and serving layer. Open weights are an upstream model property, not an open or portable Workers AI stack.

Reasoning & sources

The model catalog includes inspectable/open upstream choices, but the hosted inference stack remains managed and is not itself open or portable. Check whether the chosen model’s weights and license permit inspection or relocation, and whether the serving runtime moves too; API compatibility alone is not an open stack.

The model catalog includes inspectable/open upstream choices, but the hosted inference stack remains managed and is not itself open or portable.

The reviewed sources do not publish the serving runtime or make its infrastructure reproducible outside Cloudflare; each model’s license and portability differ.

Worker treatment

↑

Cloudflare’s impact report discusses company and supplier responsibility at a broad level, but the reviewed materials do not disclose worker-level outcomes for Workers AI model data, moderation, or data-center supply chains.

Reasoning & sources

Corporate responsibility reporting is not a service-specific audit of the people whose work supports Workers AI. Ask for supplier and contractor coverage, worker grievance channels, and remediation outcomes linked to the actual infrastructure and models used.

No relevant labor audit, annotation pay disclosure, or worker-outcome report was identified in the reviewed sources.

Government & military

↑

Cloudflare advertises defense and intelligence services, but its defense page does not establish that Workers AI is included in a named military contract or deployment.

Reasoning & sources

The company’s defense-sector offering is relevant commercial context, but it is not endpoint-level evidence for Workers AI. Keep this product-specific relationship unrated until the award and model/API scope are disclosed; do not transfer a general defense relationship to every Cloudflare product.

This is a bounded source review, not evidence that no undisclosed government customer exists.

BDS status

↑

We did not find Cloudflare, Inc. among the named targets in the two official BNC campaign lists. This is a limited list check, not BDS approval.

Reasoning & sources

Cloudflare, Inc. is not a named target in the checked lists, but BDS uses selective campaigns, so absence cannot establish a clean record. Your selected model, cloud and billing route can introduce other companies into the purchase.

This review covers the named targets in the BNC consumer/pressure guide and No Tech campaign, not every BDS campaign, local group, investment or supplier. Lists and ownership can change. A company’s absence from these lists is not an endorsement or proof that it has no relevant relationships.

All sources