Nvidia to Sell ‘Repackaged’ A100 Chips in China, Bypassing US Export Restrictions

A sign is posted in front of the Nvidia headquarters in Santa Clara, Calif., on May 10, 2018. (Justin Sullivan/Getty Images)

Chipmaker Nvidia plans to sell an alternative version of its A100 chip in China in a bid to comply with export restrictions set by the United States, a decision some experts see as a repackaging effort.

On Aug. 26, the U.S. government informed Nvidia that its data center chip A100 was added to the export control list issued by the Department of Commerce. As a result, the company was prohibited from selling the processor to Chinese customers without securing approval from Washington. The A100 chips enable artificial intelligence (AI) developers to speed up research and allow them to develop more advanced AI models.

In a statement to Reuters, an Nvidia spokesperson said the company’s A800 GPU [graphics processing unit] is an alternative to the A100 for customers in China and meets the U.S. government’s export control restrictions. The A800 went into production during the third quarter.

“The A800 looks to be a repackaged A100 GPU designed to avoid the recent Commerce Department trade restrictions,” Wayne Lam, an analyst at CCS Insight, told the media outlet.

“China is a significant market for Nvidia, and it makes ample business sense to reconfigure your product to avoid trade restrictions.”

The U.S. government implemented export restrictions on chip sales to address the risk of the items being used for military purposes. Chips such as the A100 allow data centers to process large amounts of information.

Washington’s rules impose a cap on the speeds at which the chips can communicate with each other, thus ensuring that China doesn’t gain access to high-end chips. Firms are prohibited from selling chips with communication speeds of 600 gigabytes per second and higher.

The chip-to-chip transfer rate of the A800 is said to be 400 gigabytes per second, a downgrade from the A100’s 600 gigabytes.

Financial Impact

Nvidia had acknowledged that limiting chip sales to China would deliver a severe financial blow to the company, given the size of the Chinese market.

The firm estimated about $400 million in potential sales to China in the fiscal third quarter to be affected by the U.S. government’s new licensing requirements. An alternative chip such as the A800 is expected to help lessen the impact of the restrictions.

Nvidia shares, which rose on Nov. 8 on the news, have fallen by more than 50 percent year to date. The company is scheduled to report its quarterly results on Nov. 16.

The importance of Nvidia’s chips in supercomputing was highlighted in an Oct. 11 report by the Washington-based think tank Center for Strategic and International Studies (CSIS).

In the report, Gregory C. Allen, a senior fellow in the Strategic Technologies Program at CSIS, said Nvidia and Advanced Micro Devices are two of the few chip designers in the world that are capable of manufacturing chips for AI or supercomputing, with very fast interconnect speeds and powerful parallel processors.

CUDA, Nvidia’s software ecosystem, is used by programmers widely to write massively parallelized software, which encompasses “basically all modern AI software,” he noted.

As such, “any customer who seeks to stop using Nvidia chips has to leave the CUDA ecosystem,“ Allen said in the report. ”[Thus], the combined offering of CUDA software and Nvidia hardware [explains] why Nvidia accounts for 95 percent of AI chip sales in China.”