Nvidia develops new B30A chip for China with half the flagship’s power

Nvidia is developing a new AI chip for China based on its latest Blackwell architecture that will outperform the H20 model currently sold there, according to sources familiar with the matter. The move comes as President Trump recently opened the door to allowing more advanced Nvidia chips in China, though regulatory approval remains uncertain amid ongoing tensions over AI technology access.

What you should know: The new chip, tentatively called the B30A, will use a single-die design delivering roughly half the computing power of Nvidia’s flagship B300 accelerator.

• A single-die design places all main integrated circuit components on one continuous piece of silicon rather than splitting them across multiple dies.
• The chip will feature high-bandwidth memory and Nvidia’s NVLink technology for fast data transmission between processors, similar to the H20.
• Nvidia hopes to deliver samples to Chinese clients for testing as early as next month, though specifications aren’t finalized.

The big picture: China generated 13% of Nvidia’s revenue in the past financial year, making access to cutting-edge AI chips one of the biggest flashpoints in U.S.-China trade tensions.

• Nvidia only received permission in July to restart H20 sales after being abruptly ordered to stop in April.
• Trump suggested last week that a new China chip might have “30% to 50% off” computing power, calling the H20 “obsolete.”
• The president announced an unprecedented deal requiring Nvidia and AMD to give the U.S. government 15% of revenue from advanced chip sales in China.

Competitive pressures: Huawei’s rapid chip development progress is forcing Nvidia to balance regulatory compliance with market retention.

• Huawei’s latest models reportedly match Nvidia in computing power, though they lag in software ecosystem support and memory bandwidth capabilities.
• Chinese state media recently alleged that Nvidia’s chips pose security risks, with authorities cautioning Chinese tech firms about H20 purchases.
• Nvidia argues retaining Chinese interest in its chips—which work with Nvidia’s software tools—prevents developers from switching completely to rivals.

Additional developments: Nvidia is simultaneously preparing a separate China-specific chip designed primarily for AI inference tasks.

• The RTX6000D will sell for less than the H20, reflecting weaker specifications and simpler manufacturing requirements.
• It features conventional GDDR memory with 1,398 gigabytes per second memory bandwidth—just below the 1.4 terabyte threshold that triggered the April H20 ban.
• Small batches are set for delivery to Chinese clients in September.

What they’re saying: “We evaluate a variety of products for our roadmap, so that we can be prepared to compete to the extent that governments allow,” Nvidia said in a statement.

• “Everything we offer is with the full approval of the applicable authorities and designed solely for beneficial commercial use.”
• U.S. legislators from both parties worry that access to even scaled-down flagship AI chips will impede American efforts to maintain AI leadership.