NVIDIA declared the dawn of the Agentic AI era by unveiling a new platform that will lead the next-generation artificial intelligence (AI) computing paradigm.

NVIDIA unveiled the 'Vera Rubin' platform at 'GTC 2026' held in San Jose, California, on the 17th (local time) and announced that it has begun mass production of seven new chips to expand the world's largest AI factory.

The Vera Rubin platform is designed to handle the entire AI process, from AI training and inference to test point scaling and agent-based decision-making, as a single integrated system.

Through this, NVIDIA's strategy is to transition its individual server-centric AI infrastructure into a massive supercomputer structure based on racks and PODs.

The platform unveiled this time is centered around the Vera CPU and Rubin GPU and consists of an NVLink 6 switch, ConnectX-9 SuperNIC, BlueField-4 DPU, Spectrum-6 Ethernet switch, and GROCK 3 LPU.

These chips operate like a single massive AI system, simultaneously supporting large-scale pre- and post-training and real-time agentic inference.

NVIDIA also unveiled a Vera Rubin-based rack-unit system.

The NVL72 GPU rack is By integrating 72 Rubin GPUs and 36 Vera CPUs, the number of GPUs was drastically reduced compared to the previous generation, while significantly improving cost per token and power efficiency.

This enables more efficient training and inference of very large mixed expert (MoE) models.

The Vera CPU rack for reinforcement learning and agentic workloads features a high-density liquid cooling structure to stably provide large-scale CPU computing environments.

In addition, the Grok 3 LPX inference acceleration rack supports ultra-low latency inference and large-scale context processing, making it optimized for the implementation of next-generation AI agents.

Changes also continued in the storage and network areas.

The BlueField-4 STX storage rack efficiently manages large-scale KV cache data for AI models, while the Spectrum-6 SPX Ethernet rack connects data flows across the AI factory at high speed.

NVIDIA plans to expand cooperation with cloud providers, system manufacturers, and AI research institutions, centering on the Vera Rubin platform.

Major global cloud companies and AI model developers expect to be able to operate larger and more complex AI models with low latency and cost by utilizing this platform.

NVIDIA stated, “AI infrastructure has now gone beyond simple computing resources to become a key factor determining national and industrial competitiveness,” adding that “Vera Rubin will set a new standard for building next-generation AI factories.”