Friday, Jul 10 | --:--
Back to home

Infrastructure

AI Infrastructure United States

NVIDIA Unveils Vera Rubin Platform for Agentic AI at GTC 2026

At the March 16, 2026, GTC keynote in San Jose, NVIDIA CEO Jensen Huang unveiled the Vera Rubin full-stack platform, a new generation of seven specialized chips and rack-scale systems purpose-built for the demands of agentic AI and large-scale AI factories.

Tech Insights Reporter 5 min read San Jose

TLDR

NVIDIA announced the Vera Rubin platform during Jensen Huang's GTC 2026 keynote, combining a new Vera CPU with Rubin GPUs, NVLink 6, ConnectX-9, BlueField-4, Spectrum-6 Ethernet, and integrated Groq 3 LPUs in production-ready rack systems optimized for long-running autonomous AI agents.

Vera Rubin Full-Stack Design

The platform features extreme co-design across seven new chips:

  • Vera CPU for sequential reasoning in agents.
  • Rubin GPU for high-throughput prefill.
  • NVLink 6 Switch for 260 TB/s all-to-all scale-up.
  • Additional components for networking, storage (BlueField-4), and switching.

Rack-scale offerings include Vera Rubin NVL72 GPU racks and dedicated Vera CPU racks. The system targets 10x better performance per watt in key workloads compared to prior generations and supports massive token generation rates needed for agentic systems.

Huang highlighted $1 trillion in visible purchase orders for Blackwell and Vera Rubin systems through 2027, doubling prior estimates.

Agentic AI Focus

The architecture prioritizes the full inference lifecycle for autonomous agents: high-bandwidth prefill on Rubin GPUs paired with low-latency, deterministic decode on Groq LPUs. Reference designs like Vera Rubin DSX AI Factory and Omniverse blueprints help customers simulate and deploy large-scale infrastructure.

Partners including hyperscalers, OEMs, and AI labs are already preparing deployments for the second half of 2026.

Why this story matters

Vera Rubin represents NVIDIA's shift from GPU-centric offerings to complete AI factory platforms tailored for the next phase of AI — persistent, multi-step agentic systems rather than one-shot inference. The $1 trillion demand signal underscores the scale of infrastructure buildout underway across the industry.

Sources

Featured Image Alt Text

Jensen Huang on stage at GTC 2026 announcing the Vera Rubin platform with rack-scale AI systems

Tags

NVIDIA, GTC 2026, Vera Rubin, Agentic AI, AI Factories, Inference, Rubin GPU

Scroll to continue reading