NVIDIA Blackwell B30 AI Chip for China Expected to Generate High Demand; Set to Lead the Inferencing Market

There’s considerable speculation about NVIDIA’s forthcoming AI chip designed for the Chinese market. Recent reports suggest that Team Green is preparing a potentially advantageous offering that aims to capture significant demand.

NVIDIA’s B30 AI Chip: A 40% Cost Reduction Compared to the H20 AI GPU

The last chip from NVIDIA tailored for the Chinese market was the H20 AI accelerator. Following US export restrictions, this model has been banned for domestic use. In light of these developments, NVIDIA is reportedly working on a new solution that, while less powerful than its predecessor, is expected to fulfill a large market demand. This information stems from leaker @Jukanlosreve, who cites industry insights suggesting that the B30 AI chip, codenamed “Blackwell, ”is anticipated to enjoy a positive reception in China.

NVIDIA B30 Industry Notes: • China’s major internet company A estimates that the performance of NVIDIA’s B30 is approximately 75% that of the H20. They placed an order for hundreds of thousands of units (USD 1 billion) in late June, with delivery expected in August.•…

— Jukan (@Jukanlosreve) July 12, 2025

It is essential to approach these rumors with caution, as NVIDIA’s operations in China can shift rapidly. Current claims suggest that the company aims to deliver millions of units by year’s end, with large technology firms in China eager to acquire these chips. In terms of performance, the B30 AI chip is projected to be 25% less efficient than the H20 GPU. This decrease in performance is attributed mainly to the anticipated shift from High Bandwidth Memory (HBM) to GDDR7 modules, alongside other technology limitations.

Despite these performance disparities, the B30 chip is expected to generate demand similar to past models. This is largely due to the Chinese market’s preference for economical inference solutions, especially since foundational model training has already been addressed by previous chips, like the Hopper H100. The B30 is particularly suited for deployment in small- to medium-sized applications, providing cost-efficient computational resources for numerous Chinese Cloud Service Providers (CSPs).

Additionally, the B30 AI chip is reported to function effectively within rack-scale solutions; an 8-card cluster could achieve impressive bandwidth levels of up to 1.2 TB/s. NVIDIA’s strategy appears to prioritize affordability over high-end specifications, making the new chip around 30% more energy-efficient compared to the H20 while maintaining approximately 40% lower procurement costs. Although specific details about the chip remain sparse, it’s evident that NVIDIA aims to engage the broad AI sector in China.

NVIDIA’s CEO is actively working on strengthening the company’s profile in China. He recently met with President Trump, who acknowledged NVIDIA’s impressive $4 trillion market valuation. Furthermore, there are speculations about an upcoming visit by Jensen to China, highlighting NVIDIA’s commitment to capitalizing on the opportunities presented by the Chinese AI landscape. This push for engagement with Chinese markets may also be linked to ongoing efforts for regulatory approvals from the Trump administration.

Source&Images