NEW YORK, Feb 16, 2026, 15:23 EST — Market closed Nvidia said in a blog post that new tests of its Blackwell Ultra platform showed up to 50 times higher inference throughput per megawatt than its prior Hopper generation, translating into as much as 35 times lower “cost per token” — the basic unit of text AI models process. The company said cloud providers including Microsoft, CoreWeave and Oracle Cloud Infrastructure are deploying its GB300 NVL72 systems for low-latency, long-context uses such as coding assistants and other “agentic” tools that can take steps to complete tasks. “As inference moves to the center of AI production, long-context performance and token efficiency become critical,” said Chen Goldberg, senior vice president of engineering
