Vai al contenuto
AI.info

tools

NVIDIA NIM

NVIDIA NIM fornisce microservizi di inferenza containerizzati e ottimizzati per distribuire modelli di IA su infrastrutture accelerate da NVIDIA.

NVIDIA NIM

In inglese

NVIDIA NIM packages AI models, optimized inference engines, runtime dependencies, and standard APIs into deployable software containers. Teams can use hosted APIs for prototyping or self-host NIM in clouds, data centers, workstations, and edge environments.

Developers use NIM to build AI agents, virtual assistants, document-processing systems, shopping tools, and 3D configurators. Development access is free through the NVIDIA Developer Program; production use requires NVIDIA AI Enterprise licensing, and the license does not cover the underlying model.

Features

  • Deploy prebuilt inference microservices with a single command
  • Expose models through industry-standard APIs
  • Run on NVIDIA-accelerated cloud, data center, workstation, and edge infrastructure
  • Support models based on TensorRT-LLM, vLLM, and SGLang
  • Run fine-tuned models with self-hosted NIM endpoints
  • Scale deployments on Kubernetes and cloud service providers
  • Use hosted APIs accelerated by DGX Cloud for prototyping
  • Integrate NIM endpoints with OpenAI-compatible client code

Use cases

  • Build AI virtual assistants for customer support and business processes
  • Process documents with generative AI
  • Create hyperpersonalized shopping experiences
  • Deploy 3D product configurator applications
  • Prototype AI agents with NVIDIA-hosted APIs
  • Run production inference inside a controlled infrastructure environment

Pros

    Cons

      Capabilities

      • Choice of models — “Unlock Enterprise-Ready Inference for Thousands of Open Models” source
      • Self-hosted — “Download and self-host NIM microservices for research and development as part of the NVIDIA Developer Program.” source
      • Runs models for you — “Get access to unlimited prototyping with hosted APIs for NIM accelerated by NVIDIA DGX™ Cloud” source
      • Builds agents and workflows — “Build impactful agentic AI applications with comprehensive reference workflows featuring NVIDIA acceleration libraries, SDKs, and NIM microservices.” source

      Get it

      Pricing

      Starting price
      $1125/yr
      Prices checked
      2026-11-05

      Subscription (self-managed)

      • $4,500 / GPU
      • $1,125 / GPU
      • $9,000 / GPU
      • $2,250 / GPU
      • $13,500 / GPU
      • $3,375 / GPU
      • $18,000 / GPU
      • $4,500 / GPU
      • $18,000 / GPU
      • $4,500 / GPU
      • Includes support
      • Available for 1-, 2-, 3-, 4-, and 5-year terms

      Perpetual (self-managed)

      • $22,500 / GPU
      • $5,625 / GPU
      • 5 years support

      Production (cloud-hosted)

      • $1 / hour / GPU + CSP Instance Cost(s)
      • Consumption / Pay as you go
      • Limited to 3 calls

      Development (cloud-hosted)

      Free

      • Free to use or BYOL + CSP Instance Cost(s)
      • Developer Forum and Discord

      Private Offer (cloud-hosted)

      Price on request

      • 1- to 3-year subscription
      • NVIDIA AI Enterprise Support
      Official website