NVIDIA CEO Jensen Huang has finally given an official confirmation that the firm’s most advanced AI chip, the NVIDIA Rubin, was being prepared at TSMC. And the executive also disclosed that NVIDIA has taped out half a dozen different Rubin chips, which include both general-purpose CPUs, accelerators, and a new silicon photonics chip, all of which are currently at TSMC’s fabs and being tested on production test equipment.
NVIDIA Rubin AI Architecture: 6 Chips Taped Out at TSMC
Green team doesn’t let up on product cycle. Just a few months after the debut of the Blackwell Ultra GB300 AI servers, NVIDIA appears well on its way with the NVIDIA Rubin AI chip project. “Yes, the Rubin architecture has been fully taped out at TSMC, with six different chips (including not just GPUs, but also: a new CPU, the NVLink Switch, and silicon photonics),” confirmed Huang. This is a sign of the massive overhaul of NVIDIA’s technology stack.
“My primary reason for visiting is to go to TSMC, because you know we’ve got a next generation architecture, called Rubin,” Huang added. “Rubin is really advanced; we have already taped out six brand‐new chips to TSMC.”
RubinAI: The Next Evolution In Computing Power
The NVIDIA Rubin AI chip is a breakthrough in computing performance, with upgrades at every layer of the stack. NVIDIA is expected to use HBM4 memory for next-gen R100 GPUs, no longer relying on HBM3E standard. Likewise, Rubin chips will also be in TSMC 3nm (N3P) process node with the advanced CoWoS-L packaging.
Even more importantly, Rubin is going to be NVIDIA’s first chiplet-based architecture with their 4x reticle design (compared to 3.3x on Blackwell). This makes the NVIDIA Rubin AI chip a generational upgrade in a similar way that Hopper was in its own era.
Market Timeline and Expectations
In terms of availability, the NVIDIA Rubin AI comes to market in 2026-2027, if everything works out well with trial production at TSMC. With such high-levels of innovation – across CPU and GPU improvements, to NVLink and photonics – the Rubin architecture is going to be one of NVIDIA’s most significant launches to date.
Driven by AI workloads, data centers and next-gen compute requirements, the NVIDIA Rubin AI chip is poised to be the cornerstone of forthcoming AI acceleration.
FAQs
NVIDIA Rubin is the next gen NVIDIA architecture with CPUs, GPUs, silicon photonics built for intensive AI workloads.
NVIDIA has taped-out six Rubin chips, including CPU, GPU, NVLink Switch, and photonics devices.
Full-scale production of the Rubin AI chip is scheduled for around 2026–2027 following the trial production at TSMC.
The NVIDIA Rubin AI chip brings in HBM4 memory, TSMC 3nm N3P process, chiplet design, 4x reticle, which is a bigger leap compared to Blackwell.
Rubin is a full-stack upgrade to keep NVIDIA at the forefront in AI, HPC and data center markets.
