NewsBite analysis · AI, Business, Hardwarewccftech

NVIDIA Rubin & Rubin Ultra Platforms Facing Design/Spec Issues As Per Rumors While AMD MI500 Positioned For 2H 2027 Launch

Rumors have surfaced regarding NVIDIA's upcoming Rubin and Rubin Ultra AI platforms, indicating significant design and specification challenges. Reports suggest that the company is grappling with HBM4 memory speed and capacity limitations, yield and warpage issues with chip packaging, and required heat spreader redesigns. These problems may lead to delays or scaled-back specifications. Specifically, the Rubin Ultra design is reportedly being reduced from a 4-die to a 2-die per GPU configuration, and HBM4E memory capacity is being cut from 1 TB to 768 GB due to yield problems at memory suppliers Micron and SK Hynix. In contrast, AMD's MI500 platform is on track for a second-half 2027 launch, featuring a 4-die layout and 12-Hi HBM4E memory. Both platforms are expected to incorporate co-package optics (silicon photonics). While these rumors are unconfirmed, NVIDIA has historically overcome similar challenges with its Blackwell generation. The developments underscore the intensifying competition in AI hardware and the increasing complexity of advanced semiconductor manufacturing.