Nvidia said late on Thursday that the recently announced U.S. export rules require the company to obtain an export license to sell high-performance graphics processors to China. On the one hand, these new policies might already cost the company some $400 million in sales this quarter. On the other hand, this may promptly prevent the completion of H100 development, which will further affect sales of the company’s data center products.
“On August 26, 2022, the U.S. government, or USG, informed Nvidia that the USG has imposed a new license requirement, effective immediately, for any future export to China (including Hong Kong) and Russia of the Company’s A100 and forthcoming H100 integrated circuits,” a statement (opens in new tab) by Nvidia reads.
In a bid to prevent Nvidia’s high-performance GPUs from being used by adversary military, the U.S. government now requires Nvidia to get an export license to sell any existing or future GPU achieving ‘both peak performance and chip-to-chip I/O performance […] that are roughly equivalent to the A100’ as well as systems on their base to China or Russia.
An export license is needed to ship actual products to clients for revenue and product samples to partners to support product development. Consequently, Nvidia may miss $400 million in data center sales expected from China in Q3 FY2023 earlier this quarter if appropriate licenses are not granted in time. Nvidia’s data center revenue totaled $3.806 billion last quarter, including compute GPUs and various solutions from Mellanox, so $400 million is a sizable sum for the company.
Furthermore, since Nvidia has numerous manufacturing partners in China, its inability to ship them samples may prevent it from finishing the development of its Hopper H100-based products (and possibly Grace Hopper-based products, too) on time, which will affect its data center revenue in Q4 FY2023 (which ends in late January 2023).
“The new license requirement may impact the Company’s ability to complete its development of H100 in a timely manner or support existing customers of A100 and may require the company to transition certain operations out of China,” the company said. “The company is engaged with the USG and is seeking exemptions for the company’s internal development and support activities.”
Nvidia’s A100 is the company’s current-generation datacenter GPU, introduced in mid-2020. Nvidia’s A100X and H100 offer higher performance than A100, so they automatically require Nvidia to get a license. Meanwhile, any cut-down version of Hopper H100 will likely still outpace A100 (unless Nvidia deliberately lowers their performance and makes them less competitive), so Nvidia will still be unable to sell them to Chinese clients without an export license. Meanwhile, Nvidia already does not sell products to Russia.
One of the ways for Nvidia to get around export requirements for data center GPUs is to sell its A30 products instead of A100 products to customers in China and Hong Kong. Of course, the A30 offers considerably lower performance than A100 and does not offer similar scalability. Still, if someone needs an Nvidia GPU for artificial intelligence, analytics, or high-performance computing, A30 is an option.
“In addition, the company is engaging with customers in China and is seeking to satisfy their planned or future purchases of the company’s data center products with products not subject to the new license requirement,” Nvidia said. “The company’s outlook for its third fiscal quarter provided on August 24, 2022 included approximately $400 million in potential sales to China which may be subject to the new license requirement if customers do not want to purchase the company’s alternative product offerings or if the USG does not grant licenses in a timely manner or denies licenses to significant customers.”