Ethical Stack.

NVIDIA API Catalog (NIM)

NVIDIA-hosted API endpoints for selected NIM inference microservices and model catalog entries.

Score calculation

(3 + 4 + 7 + 3) ÷ 4 = 4.3

4 of six dimensions rated. Unrated dimensions are excluded from the average.

Full scoring method ↗
Setup details

NIM endpoints explicitly hosted by NVIDIA in the API Catalog; self-hosted NIM containers and third-party model terms are outside the same service boundary.

What fits your needs

NIM endpoints explicitly hosted by NVIDIA in the API Catalog; self-hosted NIM containers and third-party model terms are outside the same service boundary.

Where it fits

Use a hosted catalog API for a quick model test, or deploy a compatible NIM container in your own infrastructure.

See the evidence
Tradeoffs to expect

Route and limits

Hosted endpoints have NVIDIA’s service boundary; self-hosted containers shift hosting and data handling to your operator.

See the evidence
How to switch

Try it

Choose a model page, obtain a developer API key, and send a small request to its endpoint.

See the evidence
Practical questions

Before relying on it

Check the selected model publisher, license, hosting mode, account terms and endpoint-specific retention.

See the evidence

Energy practices

3/10↑

Public docs establish the NVIDIA-hosted NIM API Catalog endpoint and selected model route; they don’t confirm its request-level physical footprint. The catalog lets users select a hosted NIM/model endpoint and request parameters; this is a model/workload choice, with no measured physical savings.

See the evidence

This practices and disclosure grade covers NVIDIA-hosted NIM API Catalog endpoint and selected model. Public product/API documentation establishes how this route operates and the listed user controls, but gives no measured per-request energy, emissions or water. A corporate electricity claim is not allocated to this route.

1 + 1 reporting + 0 electricity + 1 controls = 3/10. This grades documented practices and transparency.

Request-level electricity, emissions, water use, serving utilization and any route-specific clean-power matching are not reported for NVIDIA-hosted NIM API Catalog endpoint and selected model.

Reporting1/3
Official documentation describes the NVIDIA-hosted NIM API Catalog endpoint and selected model operating boundary, but does not publish a dated quantitative energy, emissions or water result for this route.
Electricity sourcing0/3
The reviewed sources do not map NVIDIA-hosted NIM API Catalog endpoint and selected model to a site-specific electricity mix or time-matched clean-power supply.
Resource controls1/3
The catalog lets users select a hosted NIM/model endpoint and request parameters; this is a model/workload choice, with no measured physical savings.

Water disclosure

Water reporting has not been assessed for this choice.

Training data

—Depends on the model↑

The catalog serves NVIDIA and third-party models; select a model and publisher before assessing the training data and license.

See the evidence

NIM is an inference runtime and API catalog, not one training dataset. NVIDIA documents third-party models in the catalog, so training data provenance and permissions depend on the model publisher and release.

Selected model and release, training data sources, licenses, post-training data, and publisher controls vary across catalog entries.

Your data & privacy

—Not enough evidence↑

The API quickstart documents NVIDIA-hosted DGX Cloud inference, but the reviewed general privacy policy doesn’t show prompt retention or training treatment for each catalog endpoint.

See the evidence

NVIDIA documents that API Catalog requests for the quickstart endpoint run on NVIDIA DGX Cloud and that a developer key is required. The reviewed general privacy policy does not give a sufficiently specific statement about prompt storage, model training, endpoint-specific retention or customer control, so no numeric privacy score is assigned.

The model-specific catalog terms, request logging, prompt retention, training use and region for the selected NIM endpoint need confirmation.

Openness

4/10↑

NIM software can be self-deployed in customer infrastructure, while the API Catalog offers a separate hosted route.

See the evidence

NVIDIA documents NIM as containerized inference microservices that customers can deploy in cloud, data center, or workstation environments. This profile is limited to hosted API Catalog endpoints, where NVIDIA controls the serving runtime.

The software provides deployable containers, but this hosted API route itself is provider-managed.

Licensing and deployability vary by model; catalog pages identify third-party models and terms separately.

Worker protections

7/10↑

NVIDIA publishes third-party supplier-audit coverage, worker-protection standards, anonymous grievance channels and specific remediation examples; evidence is concentrated in manufacturing suppliers, not AI annotation labor.

See the evidence

NVIDIA’s FY2025 report says third-party RBA Validated Assessment audits covered 48% of strategic suppliers during FY25 and 91% over two years, with corrective-action engagement for all strategic suppliers. Its May 2026 forced-labor statement reports remediating discovered hiring fees, passport/document retention, working hours and penalties for leaving, including worker fee refunds and changed supplier practices. NVIDIA also offers a third-party-hosted anonymous Speak Up line to employees, suppliers and external stakeholders. These are concrete controls and company-reported remediation; they do not disclose annotator conditions or independent worker outcomes across all suppliers. These findings concern the developer or operator. This product’s use in a particular deployment needs separate evidence.

1 + 2 pay & conditions + 2 worker voice & remedy + 2 evidence & follow-through = 7/10. This grades documented worker safeguards, not a measured rate of mistreatment.

Audit figures cover strategic manufacturing suppliers, not annotation contractors or all suppliers; remediation details are self-reported, and no worker outcome evaluation independent of NVIDIA was located.

Pay & conditions2/3
NVIDIA’s standards and audits cover manufacturing suppliers and its current statement describes concrete working-hour, recruitment-fee and document-retention remediation; no AI data-worker conditions are disclosed.
Worker voice & remedy2/3
NVIDIA describes an anonymous third-party Speak Up line available to employees, suppliers and external stakeholders and non-retaliation protections; public worker usage/outcomes are not reported.
Evidence & follow-through2/3
Third-party RBA VAP audit coverage and corrective-action closure are reported for strategic suppliers, with disclosed remediation examples; outcome reporting remains company-authored and scope-limited.
Workers rubric ↗

Military ties

3/10Broad defense ties↑

NVIDIA sells government-ready AI infrastructure and publicly supports federal, defense, intelligence, and national-security work, including mission-critical AI, autonomous machines, tactical-edge systems, and classified environments. The reviewed sources do not identify a specific NVIDIA system performing targeting or combat intelligence.

See the evidence

NVIDIA's federal materials describe AI for defense and intelligence agencies, tactical-edge deployment, radar and signal processing, autonomous machines including drones, and mission-critical AI across classified environments. This is broad defense enablement, capped at 3/10. NVIDIA publishes security/compliance controls, but the sources reviewed disclose no binding restrictions on weaponization or surveillance and no independent military-use oversight outcomes. These findings concern the developer or operator. This product’s use in a particular deployment needs separate evidence.

1 + 0 use restrictions + 1 enforcement & oversight + 1 relationship disclosure = 3/10. Known defense ties lower this rating; safeguards can improve it within the relationship’s cap.

NVIDIA's component hardware and software are used in many downstream systems; specific model deployments, end-use restrictions, actual weapon targeting, and independent outcomes are not established by these sources.

Use restrictions0/4
The reviewed government materials promote broad defense and autonomous-machine uses and contain no relevant weapons or harmful-surveillance prohibition.
Enforcement & oversight1/3
NVIDIA documents government software security and compliance processes, but these assure technical security and don’t confirm independent review of military use or published remedies.
Relationship disclosure1/2
NVIDIA identifies government customers, defense/intelligence use cases, and mission capabilities, but not specific military contracts or downstream operational outcomes.

NVIDIA openly markets platforms for defense, intelligence, tactical-edge operations, autonomous machines, and classified government missions, but the reviewed evidence doesn’t confirm a specific NVIDIA product carrying out targeting or combat intelligence. The relationship limits this grade to 3/10.

Military rubric ↗

BDS targets

Not listed↑

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

See the evidence

NVIDIA is not a named target in the checked lists, but BDS uses selective campaigns, so absence cannot establish a clean record. Running downloaded weights locally is a different decision from paying the developer or a cloud host; check the host as well. These findings concern the developer or operator. This product’s use in a particular deployment needs separate evidence.

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.

More from NVIDIA

All sources

Side by side