The American technology company NVIDIA has unveiled its new Vera Rubin platform at the GTC AI Conference, a significant expansion of its AI infrastructure designed to power the next generation of autonomous “agentic” AI. The platform consists of seven new chips in full production, including the Rubin GPU, Vera CPU, and the newly integrated Groq 3 LPU. These components are engineered to work as a unified supercomputer capable of handling every stage of AI development, from massive-scale pretraining to real-time inference.
The Vera Rubin architecture moves away from standalone servers toward fully integrated systems. The NVL72 rack-scale system demonstrates major efficiency gains, training large models with a quarter of the hardware required by previous generations while offering a tenfold increase in inference throughput per watt.
These shifts aim to drastically reduce the cost per token and energy consumption for the world’s largest AI factories.
In the announcement made by NVIDIA, they have shared that major industry leaders, including OpenAI, Anthropic, and Mistral AI, have already signaled their support for the platform to scale their increasingly complex reasoning models. The hardware is also designed to be grid-flexible, utilizing new software to unlock stranded power and improve data center resiliency. Vera Rubin-based products are expected to become available through global cloud providers and system manufacturers starting in the second half of this year.
