Nvidia Unveils Vera Rubin Platform and Groq LPU Integration at GTC 2026

Nvidia announced major AI infrastructure advancements at GTC 2026, including the Vera Rubin platform with up to 35x higher inference throughput per megawatt and the integration of Groq's LPU to solve the decode bottleneck in large language models.

Details

Nvidia's Vera Rubin platform is a full-stack computing platform that includes next-generation accelerators, CPUs, and interconnects designed to dramatically improve the economics of running AI infrastructure. The headline feature is up to 35x higher inference throughput per megawatt compared to the previous Blackwell platform, as well as up to 10x more revenue opportunity for trillion-parameter models at one-tenth the cost per token. This is enabled by the integration of 7 new chips, including the Rubin GPU, Vera CPU, and Groq 3 LPU. The Groq LPU in particular solves the decode bottleneck in current GPU architectures, using a deterministic dataflow architecture with massive on-chip SRAM to eliminate the bandwidth limitation during the output token generation phase of large language model inference. Nvidia's Dynamo software layer unifies the Vera Rubin GPU and Groq LPU, allowing developers to transparently leverage the optimal hardware for each phase of the inference process.