NVIDIA has informed customers in China that it is delaying the launch of its new H20 artificial intelligence (AI) chip until the first quarter of 2024. According to two sources (via Reuters), NVIDIA said the H20 launch would be delayed until February or March because server manufacturers are having integration issues with the chip.
The HGX H20 is the most powerful of three China-focused chips NVIDIA developed to comply with new US export restrictions. As reported by industry sources, it had been slated for a November launch.
In addition to the H20, NVIDIA’s compliance chips include the L20 and L2. By opting for an AD102 GPU, the L20 series reportedly offers lower memory sizes of 48GB and 24GB GDDR6 while staying within the limited maximum compute performance of 4800 TOPS over the PCIe interface.
The new chips were designed to allow NVIDIA to retain its market share in China after being barred from shipping advanced A800 and H800 products under tightened US rules. While including most of the newest NVIDIA AI features, the compliance chips have reduced computing capabilities per export regulations.
The export curbs have enabled rivals like Huawei to win orders that may have otherwise gone to NVIDIA. According to prior reporting, Baidu recently placed a large Huawei AI chip order.
On the other hand, AMD’s CEO Lisa Su announced in August the company’s plans to develop an AI chip designed explicitly for the Chinese market, aiming to capitalize on the growing demand for AI chips in China. The MI300 accelerator is hoped to compete with NVIDIA’s dominance in the AI market.
China has become a significant market for AI chips from US companies, as only a few domestic alternatives exist currently. AMD likely wants to capitalize on this demand while tailoring its products to meet export control requirements.