Nvidia developing new AI chip for China, B30A, beyond H20

Indianexpress

Nvidia, the world’s leading designer of artificial intelligence chips, is reportedly developing a new, more powerful AI chip specifically for the Chinese market. This move comes despite ongoing geopolitical tensions and stringent U.S. export controls aimed at restricting China’s access to cutting-edge technology.

The new chip, tentatively dubbed the B30A, is said to be based on Nvidia’s advanced Blackwell architecture, representing a significant leap over the H20 model, which is currently the most powerful Nvidia chip legally permitted for sale in China. While the H20 uses the company’s older Hopper architecture, the B30A is designed with a single-die configuration. This means its core components are integrated onto a single piece of silicon, a design choice that is expected to deliver roughly half the raw computing power of the B300, Nvidia’s flagship dual-die accelerator card. The B30A will, however, incorporate high-bandwidth memory and Nvidia’s proprietary NVLink technology, features crucial for rapid data transmission between processors, similar to the H20.

Sources familiar with the matter indicate that the specifications for the B30A are still being finalized, but Nvidia aims to provide samples to Chinese clients for testing as early as next month. This development unfolds against a backdrop of complex U.S.-Sino trade relations, where access to advanced AI chips remains a central point of contention. Despite recent suggestions from U.S. President Donald Trump that a scaled-down version of next-generation Nvidia chips might be allowed into China – contingent on a deal where Nvidia and rival AMD would share 15% of their revenue from certain advanced chip sales in China with the U.S. government – regulatory approval for the B30A is far from assured. Deep-seated concerns in Washington persist about granting China excessive access to advanced artificial intelligence capabilities.

U.S. lawmakers, across the political spectrum, express apprehension that even restricted versions of advanced AI chips could undermine American efforts to maintain its lead in artificial intelligence. Conversely, Nvidia argues that retaining a presence in the Chinese market is crucial. The company believes it is vital to keep Chinese developers engaged with its chips and software ecosystem to prevent them from fully migrating to domestic alternatives, such as those offered by Huawei. Huawei has made significant strides in chip development, with its latest models reportedly matching Nvidia in certain aspects of computing power, though analysts note it still lags in critical areas like software ecosystem support and memory bandwidth.

Nvidia’s efforts to navigate this delicate balance are further complicated by recent allegations from Chinese state media, which have suggested that U.S. chips could pose security risks. Chinese authorities have reportedly cautioned domestic tech firms against purchasing the H20, a claim Nvidia refutes, asserting its chips carry no “backdoor” risks. The H20 itself has had a turbulent journey; developed specifically for China following 2023 export restrictions, its sales were abruptly halted in April of this year, only for Nvidia to receive permission to recommence sales in July.

Beyond the B30A, Nvidia is also preparing to ship another new China-specific chip, the RTX6000D, also based on its Blackwell architecture. This chip is primarily designed for AI inference tasks, which involve applying AI models to real-world data. The RTX6000D is expected to be priced lower than the H20, reflecting its more modest specifications and simpler manufacturing requirements. Crucially, its design carefully adheres to U.S. government thresholds, featuring conventional GDDR memory and a memory bandwidth of 1,398 gigabytes per second, just below the 1.4 terabyte threshold established by the April restrictions that previously led to the H20 sales halt. Small batches of the RTX6000D are slated for delivery to Chinese clients in September, further illustrating Nvidia’s strategic maneuvering to maintain its market share while complying with evolving export controls.