Nvidia has introduced Blackwell Ultra, a new AI chip designed to improve training and inference performance while keeping power consumption roughly in line with the company’s current Blackwell platform. The company says the upgrade can deliver up to 30% better performance in some workloads.

The launch underscores how aggressively Nvidia is pushing to maintain its lead in the AI hardware market as demand for faster and more efficient chips continues to rise. Performance gains without a major energy penalty are especially important for cloud providers and large companies building or running AI systems at scale.

Nvidia has spent the past several years expanding beyond graphics processors into a central supplier for the global AI boom. Blackwell Ultra appears aimed at reinforcing that position by offering a higher-output option for customers looking to accelerate model training and deployment.

The company did not frame the announcement as a full redesign, but rather as a stronger version of its existing architecture. That approach suggests Nvidia is focusing on iterative improvements that can quickly reach data centers and keep pace with intense competition across the semiconductor industry.