In This Story
China’s artificial intelligence development is being slowed by U.S. trade restrictions on advanced chips — and the gap between the two countries could “expand,” experts said.
Despite strong demand for Nvidia’s H20 chip, which was designed by the U.S.-based chipmaker to not require an export control license, trade restrictions are slowing China’s ability to improve its large language models, or LLMs, which power AI technology such as ChatGPT, experts said during a call with Jeffries analysts, according to a note. Therefore, “China’s gap with the US would at least stay, if not expand,” analysts wrote in the note.
The H20 chip has lower computing power than Nvidia’s chips sold to U.S. tech companies, and Chinese companies are turning to Huawei’s Ascend chip as a longterm solution, experts said, adding that the support will also help the Chinese tech giant improve its chip. However, experts on the call said they are worried if China has enough capacity for the advanced 7-nanometer process to meet demand in the longterm.
Huawei was reportedly having a hard time increasing production of its Ascend 910B chip — China’s best alternative to Nvidia’s chips, which are not allowed to be sold to Chinese customers — due to components in repurposed chip fabrication machines breaking down.
U.S. officials have said a different chip, the Kirin 9000s, which uses advanced 7-nanometer processing technology and powers Huawei’s Mate 60 Pro smartphone released last year, is not as advanced as chips being developed in the U.S.
“What it tells me is the export controls are working because that chip is not nearly as good, ... it’s years behind what we have in the United States,” Secretary of Commerce Gina Raimondo said. “We have the most sophisticated semiconductors in the world. China doesn’t.”
However, while the U.S. has recently cut federal research and development across industries, China has boosted research and development by 10%, Arati Prabhakar, director of the White House Office of Science and Technology Policy, said in an interview with The Verge this month. As cuts happen during the AI boom, Prabhakar said the U.S. “should be doubling down,” and that it’s “doing the work to get back on track.”
Meanwhile, experts said demand for computing power will continue to grow as companies develop larger, more advanced models, and that China’s AI industry will likely have to consolidate since there are “too many players, and both capital and computing power are in short supply.”
The shortage could impact China’s competitiveness, experts said. While some experts said Chinese companies should focus on optimizing models to perform on a competitive scale with less computing power, another expert said the early stages of AI model development need the most computing power to experiment, and optimization would only be effective when models are mature.