Nvidia Rubin AI platform unveiled as Blackwell successor and will be available later this year

Nvidia has just pulled back the curtain on its next big architectural shift, which is Rubin. Named after astronomer Vera Rubin, this six-chip platform is designed to take over where Blackwell leaves off. Rubin focuses on high-speed networking and more efficient AI processing for massive data centers. This article will discuss everything about the latest Nvidia Rubin, including its rumours, specifications, and availability. Let's get into the details. 

Nvidia Rubin Overview

The Six-Chip Architecture

Rubin is a full ecosystem of hardware. The lineup includes the new Vera CPU, the Rubin GPU, and several networking components like the ConnectX-9 SuperNIC and BlueField-4 DPU. But the point of this ecosystem for Nvidia is that they want to move toward a highly integrated setup where the compute, networking, and memory all work together to handle AI training and inference at a scale that Blackwell wasn't originally designed for.

Efficiency and Cost per Token

One of the biggest claims Nvidia is making is that Rubin will be significantly more efficient than Blackwell. They're suggesting that companies will actually need fewer GPUs to handle the same workload they're doing now. The goal, as stated by NVIDIA, here is to lower the “cost per token.”

• Also, read

• Motorola Signature enters the market with Snapdradgon 8 Gen 5 SoC and 3X periscope zoom lens

• Dell Reintroduces the XPS 16, Redefining Premium Laptops at CES 2026!

Rubin CPX: The Cost-Optimized Variant

Nvidia is also introducing a specific version called Rubin CPX. And unlike the high-end chips that use complex CoWoS packaging, the CPX uses a simpler monolithic die and GDDR7 memory. It’s estimated to cost only about 25% of a standard Rubin GPU. This is part of a "disaggregated" design where the CPX handles the heavy lifting of starting a task (prefill), while the standard GPUs handle the memory-heavy part of finishing it (decode). 

Nvidia Rubin Availability

Since this is enterprise-grade data center hardware, you won't be buying it at a retail store. The platform is expected to become available to partners like AWS and Microsoft in the second half of 2026.