N

NVIDIA Blackwell

Hardware

by NVIDIA

Data-centre GPU architecture used for training and serving large models, sold in servers and rack systems.

Find alternatives to NVIDIA Blackwell

Sold through system builders and clouds rather than priced on nvidia.com; the architecture page's only pricing route is contact sales.

What it does

• Accelerator silicon
• Model inference
• Model training

How it compares

• AMD Instinct

  Ships data-centre training and inference accelerators to the same buyers; a direct head-to-head on rack-scale AI systems, named as such in AMD's own 10-K.

  In the Data Center segment, AMD competes primarily against Intel Corporation (Intel) and Nvidia Corporation (Nvidia) with CPU, GPU, DPU and AI NIC server products.

  Regulatory filing · 25 Aug 2026
• Intel Gaudi AI Accelerator

  Overlaps on data-centre AI accelerators, and each names the other in its own filing; Intel's same 10-K also announces an x86 co-development partnership, so the two are rivals in one market and partners in another.

  Our current competitors include: • suppliers and licensors of hardware and software for discrete and integrated GPUs, custom chips and other accelerated computing solutions, including solutions offered for AI, such as Advanced Micro Devices, Inc., or AMD, Huawei Technologies Co. Ltd., or Huawei, and Intel Corporation, or Intel

  Regulatory filing · 25 Aug 2026
• AWS Trainium

  Named in NVIDIA's own 10-K among cloud companies designing competing AI silicon; the rivalry is in the accelerator, not in the cloud platform Trainium runs on.

  large cloud services companies with internal teams designing hardware and software that incorporate accelerated or AI computing functionality as part of their internal solutions or platforms, such as Alibaba Group, Alphabet Inc., Amazon, Inc., or Amazon, Baidu, Inc., Huawei, and Microsoft Corporation, or Microsoft

  Regulatory filing · 25 Aug 2026
• Atlas

  suppliers and licensors of hardware and software for discrete and integrated GPUs, custom chips and other accelerated computing solutions, including solutions offered for AI, such as Advanced Micro Devices, Inc., or AMD, Huawei Technologies Co. Ltd., or Huawei, and Intel Corporation, or Intel

  Regulatory filing · 18 Sep 2026
• Cerebras WSE-3 Turbo (WSE-3T)

  Remarkably, the Cerebras Wafer Scale Engine 3, launched in 2024, still outperforms NVIDIA’s latest Blackwell generation by almost 5x—underscoring the enduring advantage of a compute architecture purpose-built for large-scale AI inference.

  Official documentation · 20 Sep 2026
• IPU (Intelligence Processing Unit)

  Graphcore produces AI chips called intelligence processor units (IPUs) that aim to take on Nvidia.

  News reporting · 19 Sep 2026
• Journey

  Horizon Robotics, a Chinese rival to Nvidia, seeks to raise over $700M

  News reporting · 19 Sep 2026
• SambaNova RDU

  NVIDIA dominates this space with roughly 85% market share through their H100 GPUs and CUDA software ecosystem.

  Research report · 19 Sep 2026
• Blackhole

  Tenstorrent was founded in 2016 and is one of the many startups in the current race to chip away at Nvidia's dominant AI chip market share.

  News reporting · 19 Sep 2026

Sources

Pricing

Official documentation · 2 Sep 2026

Verified 2026-09-02: URL 200, name and product unchanged, no price published.

What it does

Official documentation · 7 Sep 2026

Fetched 2026-09-07: HTTP 200, <title> “The Engine Behind AI Factories | NVIDIA Blackwell Architecture”. The page states Blackwell is built to “accelerate inference and training for large language models (LLMs)” and refers to “confidential AI training, inference, and federated learning”. The record carried accelerator-silicon alone.

Sold within

Not found · 30 Sep 2026

The page presents Blackwell as "The engine behind AI factories for the age of AI reasoning—now in full production." and lists "NVIDIA Blackwell Products" it ships inside (GB300 NVL72, DGX SuperPOD, RTX PRO, DGX Station, DGX Spark, HGX B300, GB200 NVL72, GB200 NVL4), noting "Cloud providers such as Microsoft, CoreWeave, and Oracle Cloud Infrastructure are deploying NVIDIA GB300 NVL72 systems". Its only buying route is a "Contact Sales" button; it names no single system or offering through which the architecture itself is bought.

Also from NVIDIA

• NVIDIA AI Enterprise Platform

  Supported software suite for building and deploying AI workloads on NVIDIA hardware.

• CUDA Toolkit Platform

  Programming toolkit and libraries for running general-purpose computation on NVIDIA GPUs.

• NVIDIA DGX Cloud Infrastructure service

  NVIDIA's AI cloud of GPU clusters and managed AI services, which NVIDIA's own page now presents as its internal environment for building its models, and which is still sold on Microsoft's Azure marketplace.

• NVIDIA DRIVE Platform

  End-to-end autonomous-vehicle platform spanning training infrastructure, simulation and safety-certified in-vehicle compute for production autonomy from L2++ to L4.

• NVIDIA Isaac Platform

  Open robotics development platform combining simulation, CUDA-accelerated perception libraries, ROS 2 integration and humanoid tooling for building AI-powered robots.

• NVIDIA NeMo Platform

  Suite of libraries and microservices covering the AI agent lifecycle - data curation, customisation, evaluation, guardrails and monitoring.

• NVIDIA NIM Model API

  Prebuilt inference microservices that package foundation models with optimised serving engines and standard APIs, deployable as hosted endpoints or self-hosted containers.

• NVIDIA Omniverse Platform

  Platform of OpenUSD-based libraries and microservices for building simulation-ready digital twins and synthetic-data environments used to train and validate physical AI and robots.

• NVIDIA Vera Rubin Hardware

  NVIDIA's data-centre platform succeeding Blackwell, pairing Rubin GPUs with Vera CPUs at rack scale for agentic AI and reasoning workloads.

• NVIDIA ACE Platform

  Suite of AI technologies (speech, intelligence, animation models and plugins) for building conversational, actionable in-game characters, mostly MIT-licensed with some components under NVIDIA's open model license.

• NVIDIA AI Workbench Platform

  Free development-environment manager for creating, customizing and collaborating on AI applications across GPU systems, with enterprise support available through an NVIDIA AI Enterprise license.

• NVIDIA BioNeMo Platform

  Development platform for AI-driven biology and drug discovery: open models, libraries, datasets and NIM microservices for the full AI lifecycle in biopharma.

• NVIDIA BlueField Platform Hardware

  Data-processing-unit (DPU) platform combining compute with accelerated networking, storage and security for AI-factory infrastructure.

• NVIDIA Broadcast Hardware

  Free application that enhances livestreams and video calls with AI noise removal, background replacement and other audio/video effects; requires a GeForce RTX GPU.

• NVIDIA DGX Spark Hardware

  Desktop AI supercomputer powered by the GB10 Grace Blackwell Superchip for local autonomous agents, running models up to 200B parameters for inference.

• NVIDIA DGX Station Hardware

  Deskside AI supercomputer powered by the GB300 Grace Blackwell Ultra Superchip, up to 20 petaFLOPS and 748GB coherent memory, supporting models up to 1T parameters.

• NVIDIA Dynamo-Triton (Triton Inference Server) Platform

  Open-source model-deployment server for TensorRT, PyTorch, ONNX, OpenVINO and other frameworks across GPUs and CPUs, with production support and stable APIs through NVIDIA AI Enterprise.

• NVIDIA HGX Platform Hardware

  Reference baseboard specification (HGX B200, B300, Vera Rubin NVL8) that system integrators and OEMs build into complete AI servers, listed in NVIDIA's Qualified System Catalog.

• NVIDIA IGX Platform Hardware

  Edge-AI developer kits (IGX Thor, IGX Thor Mini, IGX Orin) for industrial and medical edge computing.

• NVIDIA Jetson Hardware

  Family of embedded AI modules and developer kits for robotics and edge AI (Thor, AGX Orin, Orin NX/Nano, Xavier, TX2/Nano series).

• NVIDIA OVX Systems Hardware

  Scalable data-centre server systems (e.g. OVX L40S with four or eight GPUs) for AI and graphics workloads, built by NVIDIA OVX partners.

• NVIDIA Quantum-X InfiniBand Hardware

  Quantum-X800 InfiniBand switching for large-scale AI clusters, part of NVIDIA's InfiniBand networking line.

• NVIDIA RAPIDS Platform

  Suite of open-source, GPU-accelerated data-science libraries (cuDF, cuML, cuGraph, cuxfilter), rebranding to "CUDA-X for Data Science."

• NVIDIA Riva Platform

  Deployable speech-AI library (ASR/TTS) for production inference, free to prototype via build.nvidia.com and licensed for production deployment through NVIDIA AI Enterprise.

• NVIDIA RTX PRO Hardware

  Professional GPU line (RTX PRO 2000 through RTX PRO 6000) for AI, graphics and simulation workloads across Blackwell, Ada Lovelace, Ampere and Turing architectures.

• NVIDIA Spectrum-X Hardware

  Ethernet networking platform (switches, ConnectX SuperNICs, BlueField DPUs, LinkX cables) purpose-built for generative-AI-scale data centres.

• NVIDIA TensorRT Platform

  Ecosystem of compilers, runtimes and optimization tools for high-performance deep-learning inference; TensorRT-LLM and Model Optimizer are free on GitHub, with commercial deployment options via NVIDIA AI Enterprise.

• NVIDIA Virtual GPU (vGPU) Hardware

  Software that virtualizes GPUs across VMs in enterprise data centres and cloud deployments, licensed through NVIDIA's vGPU License and Support Portal.

Open NVIDIA in the directory

Something wrong here? Send a correction — quote this product id: nvidia-blackwell.