By Yelin Mo and Fanny Potkin
BEIJING/SINGAPORE (Reuters) – U.S. chipmaker Nvidia plans to begin mass production in the second quarter of 2024 of an artificial intelligence (AI) chip it designed for China to comply with U.S. export rules, two people familiar with the matter said on Monday.
The H20 chip is the most powerful of three China-focused chips Nvidia developed to meet restrictions announced in October.
It was originally scheduled for launch last November but that plan was delayed, with sources telling Reuters at the time that the delay was due to issues server manufacturers were having in integrating the chip.
One of the people said initial production volume will be limited, with Nvidia set to primarily fulfil orders for major customers.
Both people declined to be identified as the information was confidential. Nvidia declined to comment.
Reuters previously reported, citing sources, that Chinese companies are reluctant to buy the downgraded H20 and are testing domestic alternatives amid fear the U.S. could again tighten restrictions. Last year search engine leader Baidu ordered AI chips from Huawei Technologies in a shift away from Nvidia, Reuters reported.
In addition to the H20, Nvidia plans two other chips that comply with the new restrictions – the L20 and L2. The chipmaker has yet to announce the sale of any of the three.
In late December, it launched a modified version of an advanced gaming chip designed to comply with the new rules.
Nvidia is betting on the chips to help preserve its market share in the country after tightening U.S. export restrictions barred it from shipping products including its advanced A800 and H800 AI chips.
The A800 and H800 themselves were introduced as alternatives for Chinese customers in November 2022 about a month after the U.S. first restricted exports of advanced microchips and equipment to China.
The H20, L20 and L2 include most of Nvidia’s newest features for AI work but with computing power cut back to comply with the new rules, according to SemiAnalysis’ analysis of the chips’ specifications.
(Reporting by Yelin Mo in Beijing and Fanny Potkin in Singapore; Editing by Brenda Goh and Christopher Cushing)