Nvidia's announcement of the Vera Rubin GPU dominated headlines this week, but the real story for businesses lies in the immediate gains being made with the current Blackwell architecture. While Rubin promises a significant leap in performance, Blackwell is getting faster right now, offering tangible benefits for enterprises in the near term.

Nvidia CEO Jensen Huang revealed at CES that the Vera Rubin GPU is projected to deliver 50 PFLOPs of NVFP4 inference and 35 PFLOPs of NVFP4 training performance. This represents a 5x and 3.5x increase, respectively, over the Blackwell architecture. However, Rubin's availability is slated for the second half of 2026, leaving a considerable gap for businesses seeking to enhance their AI capabilities.

The market impact of this timeline is significant. Companies investing in AI infrastructure need solutions today, not in two years. Blackwell, launched in 2024 as the successor to Hopper, is the current workhorse. Nvidia's strategy involves not only developing new architectures but also maximizing the performance of existing ones. This approach provides a continuous stream of improvements for Blackwell users, allowing them to optimize their AI workloads and gain a competitive edge.

Nvidia has a history of refining its existing architectures. As Dave Salvator, director of accelerated computing products at Nvidia, stated, the company continues to optimize its inference and training stacks for Blackwell. This ongoing optimization translates to increased efficiency and performance for businesses leveraging Blackwell GPUs.

Looking ahead, the focus for many enterprises will be on maximizing the potential of Blackwell while preparing for the eventual arrival of Vera Rubin. The key takeaway is that AI progress is not solely about future promises; it's about the continuous evolution and optimization of current technologies. Nvidia's commitment to Blackwell ensures that businesses can continue to push the boundaries of AI today, while anticipating the even greater capabilities of Vera Rubin in the future.