NVIDIA B30A China AI Chip Set for Major Blackwell Architecture Upgrades

NVIDIA is developing a new AI chip for the Chinese market. The chip is called the B30A. It will replace the current H20 AI accelerator.

This move aims to maintain NVIDIA’s dominance in China’s lucrative AI sector. New US export restrictions have limited the performance of chips sold there.

B30A Chip Leverages Advanced Chiplet Design for Performance Boost

The B30A will feature high-bandwidth HBM3E memory. This is according to a recent analysis by GF Securities. The chip will be built on TSMC’s advanced N4P manufacturing process.

It will also include NVLink interconnect technology. This allows multiple chips to work together efficiently. The real performance gain comes from its new architecture.

The B30A will use a dual-die chiplet configuration. This is a significant shift from the H20’s single-die design. This design mirrors NVIDIA’s flagship Blackwell B200 and B300 data center GPUs.

New Chip Aims to Counter Local Competition and Export Rules

The Blackwell architecture offers a substantial performance leap. This should make the B30A highly attractive to Chinese clients. NVIDIA needs this edge against rising domestic competition.

Chinese tech firms are developing their own AI accelerators. Companies like Huawei are pushing to replace NVIDIA’s technology. The B30A is a strategic product to counter this trend.

Analysts suggest a launch could happen by the fourth quarter of 2025. This depends on receiving necessary regulatory approvals from the US government. NVIDIA must navigate complex trade policies.

The new NVIDIA B30A chip represents a critical play for the AI market. Its success hinges on both technological prowess and geopolitics. This chip could define NVIDIA’s future in China.

Info at your fingertips

What is the NVIDIA B30A chip?

The B30A is NVIDIA’s next-generation AI accelerator designed specifically for the Chinese market. It is based on the new Blackwell architecture and features a chiplet design.

How does the B30A differ from the current H20 chip?

The key difference is the move to a dual-die chiplet design from a monolithic one. This, combined with the Blackwell architecture, is expected to deliver a significant performance increase over the H20.

When will the B30A chip be released?

Analysis from GF Securities points to a potential release in Q4 of 2025. This timeline is contingent on NVIDIA securing export licenses from US regulators.

Why is NVIDIA making chips just for China?

US export controls limit the performance of AI chips that can be sold in China. NVIDIA creates specific chips like the B30A to comply with these rules while still offering a competitive product.

What memory will the B30A use?

The chip is expected to utilize high-performance HBM3E memory. This provides the necessary bandwidth for demanding AI and compute workloads.

Trusted Sources: Reuters, GF Securities.