Nvidia Corp. is expanding its offerings of smarter AI models, physical intelligence for robotics and powerful enterprise AI servers. Nvidia unveiled that the Nvidia RTX Pro 6000 Blackwell Server Edition GPU, a graphics processing unit designed for servers, is now coming to enterprise servers. This new addition will allow organizations to run large language models at high speed and these 2U form-factor rack-mountable servers will use the Blackwell architecture to deliver high-performance AI inference workloads. The new Blackwell RTX Pro Servers bring GPU acceleration to traditional CPU-based workloads — including data analytics, simulation, video processing and graphics rendering — enabling up to 45 times better performance. According to Nvidia, this results in 18 times higher energy efficiency and significantly lower cost compared with CPU-only systems. Nvidia announced an expansion of its Nemotron model family, introducing two new models with advanced reasoning capabilities for building smarter AI agents: Nemotron Nano 2 and Llama Nemotron Super 1.5. These models deliver high accuracy for their size categories in areas such as scientific reasoning, coding, tool use, instruction following and chat. Designed to empower agents with deeper cognitive abilities, the models help AI systems explore options, weigh decisions and deliver results within defined constraints. Nemotron Nano 2 achieves up to six times higher token generation throughput compared to other models in its class. Llama Nemotron Super 1.5 offers top-tier performance and leads in reasoning accuracy, making it suitable for handling complex enterprise tasks. Nvidia announced Cosmos Reason, a new open, customizable 7 billion-parameter reasoning VLM for physical AI vision agents and robotics. It allows robots and vision agents to think about what they see similar to humans and plan about what’s in a scene using intelligence such as physics knowledge and common sense from training data. The company said it can help automate the curation and annotation of large, diverse training datasets, accelerating the development of high-accuracy AI models. It added that it can also serve as a sophisticated reasoning engine for robot planning, parsing complex instructions into steps for VLA models, even in new environments.