Nvidia plans to launch the Ruby architecture for GPUs in 2026. However, we can also expect some news from Nvidia next year.

Computex started today and Nvidia cannot be missed. During his keynote, CEO Jensen Huang explained his company’s roadmap for the coming years. Nvidia is the sole conductor of the AI ​​orchestra and wants to remain so in the coming years. With new architectures, it wants to make its GPUs for data centers more powerful and capable.

The Hopper generation still reigns supreme, but Blackwell will take over later this year. Huang previously announced that Nvidia plans to release new chips every year from now on. That briefly suggested that the Rubin architecture could be officially launched as early as 2025, but Huang puts an end to those rumors, confirming that Rubio won’t be released until 2026 at the earliest.

Blackwell Ultra

But don’t think Nvidia is planning a break in 2025. Huang says the GPU specialist will launch next year with Blackwell Ultra, an even more powerful version of the Blackwell B200. The Ultra version will support up to twelve HBM3e stacks and further expand the GPU’s memory, Huang said, without going too deep into technical details.

Nvidia isn’t in a hurry to move Rubin up a year, either. When the Blackwell GPUs become available later this year, they’ll be unrivaled. The competition isn’t standing still, but it would be a tough sell if another manufacturer came out with a more powerful chip this year. The addition of an Ultra version gives Blackwell a little more stretch.

Super Chip

Nvidia has a similar schedule for the Rubin generation. The first GPUs on this architecture could go into production at the end of next year and then hit the market in spring 2026. In 2027, Rubin will then receive a refresh. At least Huang also gives a small teaser for the Vera Rubin superchip, which glues together a CPU and two GPUs.

AMD is also unveiling new desktop and laptop processors at Computex. Not surprisingly, AI was the common thread in the announcement.