At GTC 2026, the annual developer conference, NVIDIA introduced Vera Rubin, its next-generation AI platform designed to succeed the Blackwell architecture and open what the company calls the “frontier of agentic AI.”
The name honors astronomer Vera Rubin, whose research helped reveal the existence of dark matter. The symbolism reflects NVIDIA’s ambition: enabling a new phase of artificial intelligence that is more autonomous, proactive, and capable of continuous reasoning.
The platform integrates the Vera CPU and Rubin GPU in a unified architecture connected through high-speed NVLink 6. The Rubin chip features 336 billion transistors, 288 GB of HBM4 memory, and delivers up to 50 petaFLOPS of compute performance, placing it among the most powerful AI accelerators ever built.
The full Vera Rubin NVL72 rack, combining 72 Rubin GPUs and 36 Vera CPUs, is estimated to cost between $3.5 million and $4 million, roughly 25% more than the Blackwell generation. Despite the price increase, demand is expected to remain strong as AI infrastructure requirements continue to scale.
NVIDIA CEO Jensen Huang positioned Vera Rubin not as a hardware iteration, but as the computational foundation for a new paradigm: “token factories.” These systems are designed to generate, manage, and orchestrate massive streams of AI reasoning autonomously.
This aligns with the shift toward agentic AI systems capable not only of responding, but planning, executing, and adapting over time.
Cloud provider CoreWeave has already signaled early adoption of Rubin GPUs as part of its expanded agreement with Meta, a strong indication of market direction.
NVIDIA’s message is clear: those who want to lead AI in 2026 and beyond will need to build on Vera Rubin.
