Nebius Token Factory

Cloud GPU and inference platform serving selected model families; model hosting and original model/data rights are separate.

Score calculation

(7) ÷ 1 = 7.0

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

Full scoring method ↗

Setup details

Nebius is infrastructure/inference provider and may make models from other developers available.

Energy

↑

Nebius documents Token Factory inference regions and infrastructure options, but the reviewed service pages provide no inference-linked power, emissions, or water measure.

Reasoning & sources

The service page helps identify where inference can run but does not attach energy or water data to a region, model, or request. Request the selected deployment’s measured or method-based footprint before comparing it with other hosts.

Request-level electricity, emissions, and water use remain unknown.

Training rights

↑

Reasoning & sources

Nebius Token Factory terms incorporate the selected model provider’s EULA and model-specific rules; Nebius’s host/privacy materials do not independently verify training-corpus rights. Model access and customer-input handling therefore must not be treated as corpus provenance. Select the model first and review its license plus corpus-provenance disclosures; prompt no-training terms do not clear the model’s original dataset.

Both Nebius-hosted customer data use and selected model corpus rights require product/model-specific terms.

Privacy

↑

Nebius says Token Factory offers a zero-retention mode in which requests and outputs are not stored or reused for training; enterprise pages also describe isolated capacity, EU/US regional deployment, and optional custom DPAs. This is an available configuration, not evidence that every default serverless request uses it; the general privacy policy does not cover processor data handled under a separate DPA.

Reasoning & sources

Nebius provides a credible privacy path through explicit zero-retention inference, regional placement, and optional DPA terms, but the protections depend on endpoint and contract. Confirm the plan, region, subprocessor list, and any batch or stateful-feature retention in the service order before sending sensitive content.

A documented zero-retention mode and regional/dedicated controls are meaningful, but the strongest protections are configuration- and contract-dependent and the public controller privacy notice excludes processor data.

The public product material does not say the zero-retention mode is the default for all endpoints; verify plan, region, subprocessors, batch/stateful features, and the applicable DPA.

Openness

↑

Nebius provides a catalog of upstream open models and lets customers upload and serve their own fine-tuned or LoRA models, but Token Factory is a managed inference stack. Model licenses and the degree to which weights can be relocated are model-specific; the service docs do not establish that the runtime itself is open.

Reasoning & sources

Nebius documents customer model uploads and access to open models, but those options do not make its proprietary serving stack inspectable or portable. Check the chosen model license and portability separately, and leave the provider-stack score unrated until runtime source or independent deployment instructions are available.

Open model availability and license depend on selected model; model catalogue and per-model cards need review.

Worker treatment

↑

Nebius’s group Code of Business Ethics and supplier-policy pages describe expected conduct, but do not publish worker-level outcomes or audit coverage for Token Factory’s model-data and infrastructure suppliers.

Reasoning & sources

These policies establish standards rather than evidence that workers in the service’s supply chain experience them. Ask for supplier coverage, worker grievance and remediation data, and independent audit results specific to the Token Factory chain before treating the commitments as verified safeguards.

Conditions for contractors, annotators, and model evaluators remain unknown.

Government & military

↑

Nebius’s Clarifai announcement expressly says the licensed inference IP excludes Clarifai’s U.S. government and defense commercial arrangements; it does not establish whether Nebius has other defense contracts or deployments.

Reasoning & sources

That exclusion narrows the scope of one transferred asset and should not be read as a company-wide no-defense commitment. Treat Token Factory’s defense relationship as unknown until a direct contract or deployment is documented, and check the model, account, and customer scope.

No inference about undisclosed customers or contracts is possible from silence.

BDS status

↑

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

Reasoning & sources

Nebius Group N.V. 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.

Nebius says its Israel AI Cloud deployment in Modi’in offers compute for local companies and research institutes. Its current Token Factory subprocessor disclosure lists Nebius Israel Ltd. for local model processing and service operations. These establish infrastructure and business presence, not a military or rights-related use.

The company disclosures do not identify military customers, surveillance use, or a conflict-related outcome. A local facility or corporate identity is not treated as harm evidence.

All sources