Nvidia is making a strategic change to its upcoming GB300 AI server platform by returning to the Bianca compute board design. This design includes 1 CPU and 2 GPUs, instead of the previously planned Cordelia board with 2 CPUs and 4 GPUs. The decision comes after signal issues were discovered in the SXM socket interface used in the Cordelia setup. According to KeyBanc analysts, this move is seen as a positive step, helping Nvidia stick to its planned Q4 2025 launch for the GB300 platform.
Reverting to the Bianca design also means Nvidia can reuse the NVL72 rack setup used in the GB200 platform. This allows for a smoother and less costly upgrade path, which should help speed up production and delivery. There are still concerns about the shipment ramp being back-end-loaded, with 70% of GB NVL racks expected in the second half of the year. However, this change provides more stability in meeting those expectations.
For Monolithic Power Systems (MPWR), the impact of this shift is mixed. The Bianca board uses a standard power configuration, similar to what was used in the GB200, where suppliers like Infineon (IFX) and Renesas already have a strong foothold. This reduces Nvidia’s need to bring in new suppliers like MPWR for this generation. However, because the Bianca design may allow for higher average selling prices (ASPs), MPWR could still gain business if it enters the mix—without hurting profit margins too much.
Overall, this redesign helps Nvidia stay on track with its ambitious AI roadmap while minimizing technical risks and ensuring better production efficiency. It also keeps existing infrastructure relevant, which can speed up the GB300 rollout and ease adoption across data centers.
Be First to Comment