Login to Continue Learning
NVIDIA’s next AI chip for China, the B30A, is set to feature HBM3E memory and performance improvements driven by architectural advancements.
NVIDIA’s B30A Chip Could Feature Serious Upgrades Over the H20, Thanks to the Chiplet Configuration
Recent speculations suggest that NVIDIA is preparing for the Chinese AI market with a new solution. The company needs to stay competitive to maintain its market dominance. According to an analysis by GF Securities, the B30A chip will be the next flagship offering in China and will share similar specifications with the H20 AI chip, such as 8-Hi HBM3E memory, TSMC’s N4P process, and NVLink interconnect. However, a significant portion of the performance boost is attributed to the Blackwell architecture.
The B30A chip will adopt a dual-die design, which differs from the monolithic design used in the H20 AI chip. This configuration aims to align the B30A with the B200/B300 in terms of performance and features. Previous estimates suggest that the B30A will have half the specifications of the B300 chip, which GF Securities has confirmed. Specifically, the B30A will feature 141 GB HBM3E 8-HI memory, TSMC’s N4P node, and NVLink with a 900 GB/s connection, similar to the H20 configuration.
NVIDIA plans to leverage the dual-die design for a performance advantage over the H20. Given the significant performance differences between the Hopper and Blackwell generations, the B30A is likely to be well-received by Chinese customers. GF Securities estimates that the Blackwell chip could debut in Q4 2025, which means Jensen and his team will need to expedite regulatory approvals from the Trump administration.
It will be interesting to see the final specifications for the B30A chip or the next solution from NVIDIA, as Chinese AI firms are actively seeking to replace NVIDIA’s tech stack. This means that Team Green must accelerate its entry into the Chinese market with a Blackwell solution.
