Supermicro, a major NVIDIA partner, has acknowledged a delay in the release of NVIDIA’s upcoming Blackwell AI GPUs, but assured customers that its liquid-cooled Hopper H200 AI GPUs will serve as a viable alternative. Supermicro CEO Charles Liang downplayed the impact of the delay on AI server makers and the market, stating that it’s a common occurrence for new technology to experience slight delays.
NVIDIA has acknowledged production yield issues with its new Blackwell AI GPUs, leading to a delay in the B100 launch. The company will re-spin the Blackwell tile and introduce a revised B200A for enterprise customers, while prioritizing the B200 for hyperscale deployments. These developments underscore the challenges associated with advanced chip design and manufacturing.
SK hynix is nearing the completion of its HBM4 memory, crucial for NVIDIA’s next-gen Rubin R100 AI GPUs. HBM4 offers significant performance and efficiency improvements, promising a leap forward in AI capabilities.
NVIDIA’s ambitious Blackwell AI GPUs are facing significant challenges due to design flaws, resulting in delays and a revised roadmap. The B200, originally slated for Q4 2024, will now see limited shipments towards the end of the year, while the redesigned B200A is expected in the second half of 2025. These delays are attributed to issues with the Blackwell die and the CoWoS-L packaging technology, leading NVIDIA to explore alternative solutions.
