Nvidia plans to commence mass production of its H20 artificial intelligence (AI) chip in Q2 2024, tailored for China to comply with U.S. export rules. The H20, part of three chips developed by Nvidia for this purpose, faces initial reluctance from Chinese companies, who are exploring domestic alternatives amid concerns of potential tightening of U.S. restrictions.
8 January 2024 – U.S. semiconductor giant Nvidia (NVDA.O) is set to commence mass production of its artificial intelligence (AI) chip, the H20, in the second quarter of 2024. This chip, tailored for the Chinese market to adhere to U.S. export regulations, is the most potent among three chips developed by Nvidia to align with restrictions imposed in October.
Originally slated for release in November, the launch faced delays attributed to integration challenges experienced by server manufacturers, as reported by Reuters. Although initial production volumes are expected to be limited, Nvidia plans to prioritize fulfilling orders from major customers. Both the H20 and its counterparts, the L20 and L2, aim to comply with the new regulations; however, Nvidia has not yet disclosed sales information for any of these chips.
Sources indicate that Chinese companies exhibit reluctance in purchasing the downgraded H20, opting to test domestic alternatives amidst concerns of potential future U.S. restrictions. Notably, last year, search engine leader Baidu (9888.HK) pivoted to sourcing AI chips from Huawei Technologies (HWT.UL), signaling a shift away from Nvidia.
In a bid to navigate tightened U.S. export restrictions, Nvidia is strategically introducing these chips to maintain its market share in China. The A800 and H800 AI chips were introduced as alternatives for Chinese customers in November 2022 after the initial U.S. restrictions on advanced microchip and equipment exports to China.
The H20, L20, and L2 chips retain most of Nvidia’s latest AI features but with reduced computing power to comply with the newly imposed regulations, according to an analysis by SemiAnalysis. – ref: Reuters