Dylan PatelFounder and Chief Analyst, SemiAnalysis · The semiconductor-AI stack

\Vera Rubin might be NVIDIA's most important chip ever. #nvidia #verarubin #SemiAnalysis

22 Mar 2026 · 1 min

Watch on YouTube

1:11

In this commentary, the speaker emphasizes Nvidia's upcoming Vera Rubin (VR200) accelerator as a potentially game‑changing chip for 2026. Built on TSMC's N3B process and equipped with 288 GB of HBM4 delivering ~22 TB/s bandwidth, the VR200 is claimed to achieve 35 petaflops of FP4 per packet. The speaker highlights Nvidia's scaling strategy: 72 VR200 chips assembled in an NVR72 rack linked via Nvidia's NVLink network, which could keep Vera Rubin at the top of performance charts. A brief comparison notes AMD's MI455X holds a memory advantage until the later‑coming Rubin Ultra, which is expected to feature a full terabyte of HBM. The speaker concludes that VR200 is the most anticipated release of 2026 and points viewers to a recent SemiAnalysis article for deeper technical details.

Chapters

Will it? Okay, I can't avoid the elephant in the room any longer. Nvidia is number one for a reason, and with Vera Rubin, Nvidia is taking the next step, moving to TSMC's N3B and HBM4. A single VR200 can crank out an insane 35 petaflops of FP4 per packet. Combined with 288 GB of ultra-fast HBM4 at 22 TB/s of bandwidth, at least that's the speed Nvidia is targeting, A single Super chip is already a force to be reckoned with, but combining 72 of these in the NVR72 rack, using Nvidia's exquisite NVLink scale-up network, Vera Rubin will very likely once again top the charts. Or will it? MI455X does have a memory advantage, at least until Rubin Ultra, which will come with a clean terabyte of HBM. No matter what, there's no doubt that VR200 is the most anticipated release for 2026. And it's also the topic of the latest Semmianalysis article, just in case you really want to know how Vera Rubin is built and how it works.