During the GTC event in San Jose, California, Nvidia introduced its upcoming Vera Rubin AI superchip, which is set to replace the Blackwell Ultra chip and is expected to be available for shipping in the latter half of 2026. Vera Rubin, much like the Grace Blackwell superchip, incorporates both a central processing unit and a graphics processing unit, with Vera serving as the CPU and Rubin as the GPU.

With upgraded features such as 4.2 times the memory of Grace and 2.4 times the memory bandwidth, along with 88 CPU cores, Vera promises twice the performance of its predecessor. Vera Rubin will also feature 288GB of high-bandwidth memory 4 (HBM4) compared to the HBM3e in the Blackwell Ultra.

Nvidia also revealed the subsequent generation of chips named Vera Rubin Ultra, slated for release in the second half of 2027. This advanced chip will combine a Vera CPU with the Rubin Ultra chip, where Rubin Ultra integrates 4 GPUs, as opposed to the 2 GPUs found in Rubin.

Similar to the Blackwell series, both Vera Rubin and Vera Rubin Ultra will be available in large server rack systems. The NVL144 system will house 144 GPUs and deliver 3.3 times the performance of the GB300 NVL72. The Vera Rubin Ultra NVL576, comprising 576 Vera Ultra GPUs with 365TB of memory, is said to be 14 times faster than the GB300 NVL72 and will be accommodated in the liquid-cooled Kyber Rack server design.

While the Blackwell chip remains an essential revenue driver for Nvidia, the company's introduction of Vera Rubin demonstrates its strategic initiatives for future systems catering to hyperscalers and organizations involved in AI services. Nvidia aims to target entities such as Amazon, Google, Meta, Microsoft, and government research facilities with its Vera Rubin series. However, concerns about overspending by hyperscalers on AI investments, President Trump's tariff threats on overseas semiconductor production, and potential export controls have impacted Nvidia's stock price.