Managed GPU inference
High-performance, reliable and scalable inference APIs backed by Dimension AI GPU infrastructure.
AI Services
Enterprise AI Inference as a Service
Serve models in production without assembling GPU serving, scaling and operations yourself. Inference is delivered from Dimension AI Factory infrastructure.
Overview
What is AI inference as a service?
Inference is the production step: a trained or fine-tuned model answering requests. Inference as a Service means Dimension AI runs the GPU serving layer — scaling, reliability and access — so applications consume model output without operating the full stack.
This sits alongside model endpoints and token services. Together they turn factory compute into consumable intelligence: APIs, hosted models and token-based consumption rather than only raw GPUs.
Capabilities
What you can consume
Model endpoints
Hosted LLMs, embedding models and other AI models accessed via API, without standing up serving for each model yourself.
Token services
Purchase and consume AI tokens through packs or subscriptions when you want intelligence as a metered resource.
Capabilities
Production requirements
Performance and scale
Serving designed for production traffic, not only laptop-scale demos, with GPU capacity that can grow with demand.
Reliability and observability
Inference is operated as part of managed AI Factory operations, including monitoring and performance optimisation.
Enterprise security
Serving can run in public, private or sovereign factory models depending on isolation and residency requirements.
FAQ
Questions teams ask
What is the difference between inference and training?
Training and fine-tuning create or adapt a model. Inference runs that model in production to generate outputs. Dimension AI supports both: GPU clusters and Bare Metal GPU for training, and Inference as a Service for serving.
Do I still need GPUaaS if I only want inference?
Not necessarily. Inference as a Service is for teams that want to consume serving. GPU as a Service is for teams that need to provision and run GPU infrastructure themselves, including training jobs.
