Nvidia's Blackwell Chips Delayed Until 2025 Due to Design Flaw

Customers who have pre-ordered Nvidia's new Blackwell flagship line of chips will experience delays due to design flaws within the B200 chips. The discovery of the flaw came late in the production process, prompting Nvidia to work on rectifying the issue with manufacturer Taiwan Semiconductor Manufacturing Company (TSMC). The flaw affects the processor die connecting two Blackwell GPUs on a single board, necessitating design rework and additional production tests before mass production can proceed.

Reports indicate that major tech companies like Meta, Microsoft, and Google have invested significant sums in acquiring these new chips for running large AI models more cost-effectively. Meta aims to expand its infrastructure, Microsoft seeks the chips for training purposes, and Google, despite having in-house custom hardware, has ordered a substantial amount of Blackwell chips. To mitigate the delays, Nvidia plans to extend the lifespan of its previous flagship Hopper chips and introduce the B200A chip, specifically designed to power lower to mid-range AI systems until the B100 and B200 chips are ready for release.

In response to the delays, Nvidia will also offer a chip compatible with the current die system, the B200A, which will be used temporarily until the design flaw is resolved. This chip is expected to transition into the B100 and B200 chips in the future. Notably, the die featured in the B200A will also be incorporated into the B20, a chip tailored for the Chinese market. Nvidia's move to develop a new chip for the Chinese market comes after U.S. export regulations limited the sale of high-powered hardware to clients in the country.