The NVIDIA Vera Rubin AI supercomputing platform was officially launched on March 16, 2026, during the company’s annual GTC 2026 conference. This new platform was introduced as a comprehensive full-stack solution designed to power agentic AI, emphasizing its role in supporting complex, multi-step autonomous AI workflows. CEO Jensen Huang presented the platform as a generational leap in AI infrastructure, highlighting its capability to handle every phase of AI development, from large-scale pre-training to real-time agentic inference.
The Vera Rubin platform integrates several key components, including the NVIDIA Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet switch, alongside the newly incorporated Groq 3 LPU. This combination of advanced hardware is engineered to function as a single, powerful AI supercomputer, providing significant performance improvements, such as up to 10x more inference throughput per watt compared to previous systems. The platform’s design focuses on optimizing for trillion-parameter models and million-token contexts, aiming to enhance the efficiency and scalability required for the next generation of AI applications.
NVIDIA positioned the Vera Rubin platform to accelerate the development and deployment of agentic AI, which involves AI systems capable of understanding context, making decisions, and executing multi-step tasks autonomously. The company has also unveiled the Vera Rubin DSX AI Factory reference design, a blueprint intended to help organizations build and operate large-scale AI infrastructure efficiently. While the Vera Rubin Space Module, a related component for orbital data centers, was also announced at GTC 2026, it did not have a specific release date at the time, with NVIDIA stating it would be available “at a later date.” However, the core Vera Rubin platform for ground-based AI factories is already in production, with its components available to partners and hyperscalers, supporting the ongoing expansion of AI capabilities. This infrastructure will enable the processing and analysis of vast datasets, crucial for complex AI operations, much like how a vector database such as Milvus handles high-dimensional vector embeddings for efficient similarity search in AI applications.