US makes it harder for chipmakers to run fabs in China amid new Alibaba processor reports – News

The U.S. Commerce Department is making it more difficult for chipmakers to operate their fabs in China, Reuters reported today.

The move affects Samsung Electronics Co. Ltd. and SK hynix Inc., the world’s two largest memory suppliers. The latter company makes much of the HBM memory in Nvidia Corp.’s graphics cards. Samsung, in turn, sells not only RAM but also processors for devices such as smartphones.

Intel Corp. is the third chipmaker affected by the Commerce Department’s new initiative. However, it’s unclear whether the move will impact the company’s business operations in practice. Intel sold its fab in China, which made memory chips, to SK hynix four years ago.

The Commerce Department will require Samsung, SK hynix and Intel to obtain licenses before installing American chipmaking equipment in Chinese fabs. The new rule is set to go into effect in 120 days. Officially will reportedly permit the chipmakers to operate their existing production lines in China, but won’t allow them to upgrade equipment or add manufacturing capacity.

The news sparked a selloff in the shares of U.S. chipmaking equipment suppliers. Lam Research Corp. ended the trading session with its stock down about 3.8%, while Applied Materials Inc. declined by 2.7%.

In 2023, the Commerce Department restricted exports of advanced chipmaking equipment to China. Companies must obtain a license before they can make such shipments. According to Reuters., U.S. suppliers are currently waiting for licenses to export billions of dollars’ worth of chipmaking hardware to China. 

In conjunction with the Reuters report, sources told the Wall Street Journal today that Alibaba Group Holding Ltd. has designed a new artificial intelligence chip. The company is reportedly using Chinese fabs to make processors. According to the Journal, those fabs use legacy foreign-made chipmaking equipment and homegrown hardware.

Alibaba’s new chip is reportedly optimized for inference workloads. It’s expected to support a broader range of use cases than the company’s earlier custom processors.

On the software side, the chip will reportedly be capable of running applications originally written for Nvidia silicon. Usually, porting software between chips requires making extensive code changes. Removing that requirement can significantly reduce the amount of time needed for the task.

Nvidia and Advanced Micro Devices Inc. recently received permission to resume AI chip exports to China. The companies will pay 15% of the revenue they generate from such sales to the U.S. government. It’s believed that Nvidia alone could receive billions of dollars’ worth of orders from customers in China by year’s end.

Photo: Unsplash