Shortest time-to-market for next-gen reasoning models and multimodal LLMs. With its massive 288GB HBM3e memory and up to 50x higher real-time inference performance, the NVIDIA GB300 NVL72 accelerates both large-scale post-training and real-time agentic reasoning, helping businesses deploy advanced AI products at 35x lower token costs.

Powering Real-Time Reasoning Agents

Larger memory capacity allows for larger batch sizing and maximum throughput performance. NVIDIA Blackwell Ultra GPUs offer 1.5x larger HBM3e memory combined with 1.5x FP4 performance compared with the predecessor, boosting AI reasoning throughput for the largest context lengths and complex thinking chains.

Driving Multimodal Innovation

Deploy high-throughput inference to energize complex workloads across audio, language, and visual domains. The NVIDIA GB300 NVL72 provides the immense memory bandwidth and density essential for scaling the next generation of AI-driven applications.

Accelerating Strategic Decisions

Transform intricate data streams into precise, actionable insights through high-throughput reasoning capabilities. The NVIDIA GB300 NVL72 empowers organizations to expedite predictive modeling and scenario-based simulations, facilitating rapid, high-confidence executive decisions across the enterprise.

Use Cases

Software and Agentic AI

Purpose-built for the "Era of AI Reasoning," providing FP4 precision support for autonomous agents capable of advanced self-correction and planning. With 288GB of HBM3e per GPU, agents can process ultra-long contexts, spanning millions of tokens, allowing for autonomous reasoning and execution across entire software architectures or massive technical libraries.

Media and Creative Industries

Drive next-generation multimodal generative applications, supporting the real-time rendering of high-fidelity 3D environments and interactive digital humans. The 8.0 TB/s per-GPU memory bandwidth eliminates throughput bottlenecks for ultra-high-definition video generation, enabling immersive entertainment and design experiences that are physically accurate and instantly responsive.

Energy and Public Sector

Deploy sovereign AI-compliant compute clusters for high-precision climate modeling and national grid optimization. The NVIDIA GB300 NVL72 delivers a significant improvement in performance-per-megawatt (AI output per unit of power) over the Hopper architecture, enabling sustainable processing of petabyte-scale disaster prediction data to strengthen the resilience of critical public infrastructure.

NVIDIA GB300 NVL72 Technical SpecsTechnical Specs