Solutions Industries Developers Partners Login
NVIDIA Blackwell · Q2 2026

Accelerated computing
for the age of AI.

From training the world's largest models to inference at the edge — the same architecture, the same software stack, from data center to laptop.

Available now

DGX B200 · Blackwell

SKU NV-DGX-B200-8GPU
STATUS
● ONLINE
FP4 Compute144PFLOPS
HBM3e1.4TB
NVLink BW14.4TB/s
14 deployments | Shipping Q2 2026 | Available in 18 regions
Why NVIDIA

The platform behind every modern AI breakthrough.

From training the largest open-weight models on DGX SuperPOD to running on RTX on a developer's laptop — one software stack, end to end.

Faster training vs prior generation
25×
Lower energy per token at inference
18
Cloud regions running DGX Cloud
3.4M
Developers building on CUDA
Capabilities

Explore our AI solutions.

Agentic AI

NVIDIA NIM microservices and NeMo blueprints to build, fine-tune, and deploy agents in production.

Learn more →

Data Science

RAPIDS, cuDF, and cuML — GPU-accelerated dataframes and ML at the speed of memory, not disk.

Learn more →

Inference

NVIDIA Triton + TensorRT-LLM serve open models at production latency on B200, H200, and L40S.

Learn more →

Conversational AI

Riva and Maxine — speech, translation, and real-time avatars deployed at telco-grade latency.

Learn more →
Latest in AI resources

Research, papers, and whitepapers.

White paper

Scaling FP4 training to 100K-GPU clusters.

How NVLink-72 and Blackwell's transformer engine compound to deliver 4× the throughput at the same wall-clock budget.

Read more →
Webinar

NIM microservices in production.

Three customer case studies on serving open-weight Llama 4 and Mixtral with TensorRT-LLM at sub-200ms p95.

Watch on demand →
Blog

Cosmos world models for robotics.

Generating photoreal sim-to-real training data with diffusion-based world models on RTX A6000.

Read article →

Ready to accelerate your AI?

Provision a B200 in DGX Cloud in minutes — no procurement cycle, no commit. Pay-per-minute pricing on the same hardware that runs the world's largest models.

Start DGX Cloud trial →
DGX Cloud pricing

Per-minute pricing on every GPU.

L40S

$1.42 / GPU·hr

For inference workloads and fine-tuning up to 70B parameters.

  • 48 GB GDDR6
  • FP8 inference
  • On-demand · spot
  • Community support

B200

$8.40 / GPU·hr

For training the largest open-weight models and exascale inference.

  • 192 GB HBM3e
  • NVLink-72 fabric
  • FP4 transformer engine
  • Dedicated solutions architect
Design system

Tokens, by NVIDIA.

Two-mode surface architecture, a single saturated green accent, 2px radius everywhere, and a corner-square that ships with every reusable card.

→ Read the full DESIGN.md

Color

grouped & named

Brand & accent

primary (NVIDIA Green)#76b900
primary-dark#5a8d00
accent-green-pale#bff230

Surface

canvas#ffffff
surface-soft#f7f7f7
surface-dark#000000
surface-elevated#1a1a1a
hairline#cccccc
hairline-strong#5e5e5e

Text

ink#000000
body#1a1a1a
mute#757575
ash#a7a7a7

Semantic & editorial

link-blue (inline)#0046a4
error#e52020
warning#df6500
accent-purple#952fc6
accent-yellow-pale#feeeb2

Typography

NVIDIA-EMEA · weights 400 / 700
display-xl 48 / 700 / 0 1.25 Accelerated computing.
display-lg 36 / 700 / 0 1.25 Explore our AI solutions.
heading-xl 24 / 700 / 0 1.25 Sub-section title.
heading-md 20 / 700 / 0 1.25 Card group title.
card-title 17 / 700 / 0 1.47 Resource card title.
body-md 16 / 400 / 0 1.5 Default body — hierarchy comes from weight, not color.
caption-md 14 / 700 / 0 1.43 BREADCRUMB · EYEBROW · BADGE

Radius

Aggressively angular — 2px everywhere
none0px · hero/footer
xs1px · accent
sm2px · default
fullcircle · avatar

Components

Buttons · tabs · inputs · cards
Pill tabs & badges
White paper Tip Preview
Text input
Focus: 2px green border replaces hairline.
Inline link

Train on NVIDIA DGX Cloud with reserved or on-demand capacity — the only blue in the system lives on inline body links like this one.

Feature card · Resource card

Feature card.

32px padding, hairline border, green corner square anchored top-left.

Learn more →

Callout stat.

Display-lg numeric in NVIDIA Green, body caption underneath.

Learn more →
Webinar

Resource card.

16:9 thumbnail at the top, body-sm description, ghost-link "Read more" — exactly the same chrome as the marketing cards above.

Watch now →
View DESIGN.md