NVIDIA has officially confirmed that its new Blackwell AI GPUs are facing production yield issues, leading to delays in the launch of its B100 model. The company will be re-spinning certain layers of the Blackwell tile to improve yields and has announced a revised version, the B200A, specifically targeted at enterprise customers. The B200, however, will remain a priority for hyperscale deployments.
Analyst firm KeyBanc had previously reported on the design flaws plaguing the Blackwell GPUs, suggesting a three-month delay in shipments. The firm also predicted that NVIDIA would effectively cancel the B100 and replace it with a lower-cost B200A, a scenario now confirmed by NVIDIA.
The B200A, born out of the redesign effort, is anticipated to be available in the second half of 2025. These issues highlight the intricate challenges associated with developing and manufacturing cutting-edge chips. The delays and adjustments demonstrate the importance of rigorous testing and optimization in the chip production process. NVIDIA’s commitment to prioritize the B200 for hyperscale customers reveals the growing demand for powerful AI infrastructure in large-scale data centers.
