Today at the Consumer Electronics Show, Nvidia CEO Jensen Huang officially unveiled the company’s new Rubin computing architecture, which he described as the cutting edge in AI hardware. The new architecture is currently in production and is expected to be more advanced in the second half of the year.
“Vera Rubin was designed to address this fundamental challenge: The amount of computation required for AI is increasing.” Huang told the audience. “Today, I can tell you that Vera Rubin is in full production.”
Rubin Architecture, the first to be announced in 2024is the latest result of Nvidia’s hardware development cycle, which has turned Nvidia into the most valuable company in the world. Rubin’s architecture will replace Blackwell’s architecture, which will then replace Hopper’s and Lovelace’s architecture.
Rubin’s chips are slated to be used by nearly every major cloud provider, including a high-profile Nvidia partnership Anthropotic, OpenAIand Amazon Web Services. The Rubin system will also be used in Blue Lion HPE supercomputer and the future Doudna supercomputer at Lawrence Berkeley National Lab.
Named for astronomer Vera Florence Cooper RubinRubin’s architecture consists of six separate chips designed to be used in concert. The Rubin GPU is located in the center, but the architecture also solves the bottleneck in storage and interacts with the new improvements in the Bluefield and NVLink systems. The architecture also includes the new Vera CPU, designed for agent considerations.
Explaining the benefits of the new storage, Nvidia’s senior director of AI infrastructure solutions Dion Harris shared the memory demands associated with modern AI system caches.
“When you start enabling new types of workflows, like agent AI or long-term tasks, that puts stress and requirements on your KV cache,” Harris told reporters on the phone, referring to the memory system is used by the AI model to condense the input. “So we’ve introduced a new level of storage that connects externally to computing devices, which allows you to expand your storage pool more efficiently.”
Techcrunch event
San Francisco
|
13-15 October 2026
As expected, the new architecture also shows significant advances in speed and power efficiency. According to Nvidia’s tests, the Rubin architecture will operate three and a half times faster than the previous Blackwell architecture in model-training tasks and five times faster in inference tasks, reaching 50 petaflops. The new platform will also support eight times more inferential computing per watt.
Rubin’s new capabilities come amid intense competition to build AI infrastructure, which has seen both AI labs and cloud providers vie for Nvidia’s chips as well as the facilities needed to power them. In an earnings call in October 2025, Huang estimated between $3 trillion and $4 trillion will be used for AI infrastructure over the next five years.