NVIDIA NIM

What it does

NVIDIA NIM supplies prebuilt, optimized inference microservices: containers that bundle an AI model, an inference engine and runtime dependencies. Applications send requests through industry-standard APIs and receive model outputs. Available models span reasoning, language, retrieval, speech, vision and safety, including NVIDIA Nemotron open models.

Teams start a container with a single docker run command, then call the endpoint with curl or the OpenAI Python client pointed at their own endpoint. NIM can also serve LLMs supported by TensorRT-LLM, vLLM or SGLang, including fine-tuned models, and it scales on Kubernetes. NVIDIA positions it between a managed API and self-hosting: the same models can be tried through hosted APIs, then run on infrastructure you control. It also ties into NVIDIA Blueprints and NeMo Relay for agentic applications.

It suits developers and enterprise AI teams that already run, or plan to run, NVIDIA-accelerated infrastructure in the cloud, a data centre, a workstation or at the edge. Free development access is offered through the NVIDIA Developer Program, but production use moves to an NVIDIA AI Enterprise licence. It is not usable without NVIDIA GPUs, and performance figures on the page come from NVIDIA's own benchmark configurations.

What it can do

• Prebuilt model inference containers with optimized engines and runtime dependencies
• Industry-standard APIs, including OpenAI-style chat and completions endpoints
• Single-command deployment of models
• Support for LLMs run with TensorRT-LLM, vLLM or SGLang, including fine-tuned models
• Kubernetes scaling across cloud, data centre, workstation and edge
• Hosted APIs for prototyping on NVIDIA DGX Cloud

Works with

• NVIDIA Run:ai
• NVIDIA Blueprints
• NVIDIA NeMo Relay
• Kubernetes

Example workflows

Self-host a model endpoint

Input: a supported model and an NVIDIA GPU host. Steps: pull and run the NIM container with docker run, then send a request to the local /v1/completions endpoint. Output: generated text served from your own infrastructure.

Connect an app via the OpenAI client

Input: a running NIM endpoint URL and key. Steps: set base_url and api_key in the OpenAI Python client, then call chat.completions with the model name. Output: chat responses inside your application.

Before you choose

• Requires NVIDIA-accelerated hardware or NVIDIA-hosted APIs; it is not a hardware-agnostic runtime
• Production use moves to a paid NVIDIA AI Enterprise licence priced per GPU
• Performance figures are NVIDIA's own benchmarks on specific configurations

Listed under

• Coding
• Sandboxes and infrastructure
• Developers
• Enterprise
• Enterprise AI
• GPU
• GPU acceleration
• Inference

Get started

Free development access; production needs AI Enterprise

1. Join the NVIDIA Developer Program and try the hosted APIs at build.nvidia.com
2. Provision NVIDIA-accelerated infrastructure and pull a NIM container
3. Point your application at the endpoint using standard API calls

Production use is via NVIDIA AI Enterprise (the pricing page covers the suite, not NIM alone). Self-managed list price is per GPU and includes support; cloud marketplace pay-as-you-go adds cloud instance costs.

Source: docs.nvidia.com/ai-enterprise/planning-resource/licensing-guide/latest/pricing.html, checked 28 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 28 Oct 2026. Each source supports part of the overview; availability and prices can change.

• NVIDIA NIM Microservices for Accelerated AI Inference | NVIDIA ↗www.nvidia.com
• Pricing — NVIDIA Enterprise Licensing Guide ↗docs.nvidia.com

Company website

nvidia.com: product, pricing and policy pages

YC directory mirror

Batch and company name

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 28 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.