Nvidia's Rubin Ultra Delayed as Integration Challenges Mount

Nvidia's Rubin Ultra AI architecture is facing integration challenges and rising component costs, which could delay its rollout to 2028.

The company's next-generation AI architecture aims to integrate more GPUs, memory, networking, and power into a single system. However, Brad Gastwirth, Senior Market Analyst from Circular Technologies, explains that this complexity is causing integration challenges. Specifically, Nvidia's move towards 144 GPUs per rack is creating new integration hurdles.

The rising costs of high-bandwidth memory (HBM) are also forcing Nvidia to test alternative configurations. As a result, smaller initial volumes may arrive in 2027, while the broader rollout is pushed towards 2028.