Nvidia confirms that its next-gen “Rubin” chips have entered trial production

Six new Nvidia chips have been taped out at TSMC

During his latest visit to Taiwan, Nvidia’s CEO, Jensen Huang, confirmed that the company has taped out six new chips at TSMC’s fabs. This includes Nvidia’s next-generation “Rubin” processor, the heart of Nvidia’s next-generation Rubin platform.

With Rubin, Nvidia’s planning to deliver more than a mere chip upgrade. With Rubin, Nvidia intends to overhaul its entire computing platform. Nvidia’s six new chips include a new CPU, GPU, Scale-Up NVLink Switch, networking chip, networking switch, and silicon photonics processor. In other words, Nvidia’s entire AI product stack is getting an upgrade.

These chips signal that Nvidia plans to deliver better rack-scale connectivity alongside higher levels of per-chip AI performance. With their upgraded AI platform, Nvidia should be able to deliver a large generational leap in AI performance. Currently, Nvidia’s Rubin platform is expected to be launched between 2026 and 2027. However, this timeline depends on how the chip’s trial production goes.

(Nvidia’s CEO Jensen Huang speaks to Taiwanese reporters)

Rubin is expected to be the first Nvidia AI product to use HBM4 memory. Furthermore, Rubin is expected to use TSMC’s 3nm technology alongside their CoWoS-L packaging technology. Previously, Nvidia has claimed that Rubin will use a chiplet design (a first for Nvidia). The chip itself is a 4x reticle design, whereas Blackwell is a 3.3x reticle design. That alone should highlight how much bigger Nvidia’s next-gen AI chips are.

Rubin will be a big launch for Nvidia, and that launch is now one step closer.

You can join the discussion on Nvidia’s CEO confirming that “Rubin” has entered trial production on the OC3D Forums.