November 7th， as reported by Kuaitech， NVIDIA’s specially designed B30A chip for the Chinese market has once again been prohibited for sale by the United States government， presenting a significant setback for NVIDIA CEO Jensen Huang.
According to U.S. media outlets， the White House has formally notified federal agencies of its decision to prevent NVIDIA from exporting its latest miniaturized artificial intelligence (AI) chip， the B30A， to China.
NVIDIA had reportedly already supplied samples of the B30A chip to several of its clients in China.
The B30A chip， when deployed in large， efficiently configured clusters， is capable of training large language models (LLMs)， a capability highly sought after by many Chinese technology companies given the rapid advancements and demand in this sector.
NVIDIA employees have indicated that the company is actively engaged in efforts to modify the B30A’s design， holding out hope that the U.S. government might reconsider its stance.
A spokesperson for NVIDIA stated that the company “has zero share in China’s highly competitive data center computing market and therefore has not included it in its performance expectations.” This statement suggests that any potential sales of the B30A were not significantly factored into NVIDIA’s financial outlook， likely due to the anticipated regulatory hurdles.
Based on previously leaked information， the B30A chip’s price was expected to be approximately double that of the H20 chip， which currently retails between $10，000 and $12，000. This would place the B30A’s potential price point between $20，000 and $24，000， equivalent to around 171，000 Chinese Yuan.
Moreover， the B30A is believed to be a modified version of the Blackwell B300A， which is based on a single-chip design. Specifications indicate that the B300A utilizes TSMC’s 4nm process， advanced CoWoS-L packaging， and is equipped with 144GB of HBM3E memory， operating at a power consumption of 600W. The B30A’s potential performance was rumored to be around 80% of the performance of the base Blackwell GPU.
