Google Vertex AI — AI Glossary

Reference details

Related terms

• AWS Bedrock
• Azure OpenAI Service
• Google Gemini

How it works

1. Models are called through a project- and region-scoped endpoint, so requests, data residency and quota are all tied to a cloud project rather than to an API key.
2. Authentication uses the cloud's own identity system, which is what lets model access be granted by role alongside other infrastructure permissions.
3. First-party and third-party models are served through the same surface, so switching between them is a model identifier rather than a new integration.
4. Network controls can confine traffic to a private perimeter, which is usually the reason an enterprise chooses a managed surface over a direct API.
5. The surrounding platform — registry, pipelines, feature store — is what makes it an ML platform rather than a model endpoint, and that is the half most comparisons omit.

Example: What the managed surface actually changes

Calling a model directly needs an API key and an outbound HTTPS route. Through a cloud surface the same call needs a project, a region, an IAM role granted to a service account, and possibly a private network path — more setup for one request, and the reason is the second request: quota, audit, residency and billing then arrive with the infrastructure rather than being rebuilt around it. The trade is a higher floor for a lower slope, which is why it rarely pays for a prototype and usually pays at organisational scale.

Commonly confused with

Comparable with Bedrock and Azure's OpenAI service: a cloud's managed surface over foundation models plus the training and MLOps tooling around them. That second half is the difference — it is a full ML platform, not only a model gateway, so it competes with SageMaker as much as with a model API. Gemini is also reachable through a separate developer API with different quotas and terms.

When to use it

Reach for it when:

• The workload already runs in that cloud and should inherit its identity, networking and billing.
• Data residency or private networking are requirements a direct API cannot satisfy.
• The surrounding ML platform — registry, pipelines, tuning — is wanted, not just inference.

Reach for something else when:

• A prototype that needs one API call, where the setup cost dominates.
• Deliberate provider independence, where a cloud-specific surface works against it.
• Only inference is needed and the platform half goes unused.

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Google Vertex AI — AI Glossary.

• Browse models
• Review providers
• Inspect benchmarks
• Explore workflows
• Learn AI concepts
• Read decision guides
• Review trust and methodology
• Start a decision path