New performance data from SemiAnalysis shows NVIDIA Blackwell Ultra delivering breakthrough advances in AI inference. The GB300 NVL72 systems now achieve up to 50x higher throughput per megawatt compared to the previous NVIDIA Hopper platform, translating to 35x lower cost per token for AI applications.

This dramatic improvement comes as AI agents and coding assistants drive explosive growth in software-programming-related queries, jumping from 11% to about 50% of AI usage last year according to OpenRouter's State of Inference report. These applications demand low latency for real-time responsiveness across multistep workflows and long context when reasoning across entire codebases.

Leading inference providers including Baseten, DeepInfra, Fireworks AI, and Together AI have already adopted the NVIDIA Blackwell platform to reduce cost per token by up to 10x. The new Blackwell Ultra platform builds on this foundation through extreme codesign across chips, system architecture, and software optimizations. These advances accelerate performance across AI workloads from agentic coding to interactive coding assistants while driving down costs at scale.