NVIDIA has halted shipments of Hopper generation AI accelerators to China despite authorization from US authorities, and the company’s interim CEO, Colette Kress, confirmed that no further shipments are currently planned.
During an earnings call with investors on Tuesday, Kress stated that during the quarter ending October 25, NVIDIA sold $2.4 billion worth of licensed Hopper generation server products to China. However, such shipments are no longer factored into forecasts for future periods.
There are no licensed Hopper generation server products in our forecast, and we are no longer shipping such products to China.
She did not explain the reasons for halting the shipments or clarify whether this decision would be permanent.
The Hopper family includes the H200, a powerful accelerator for training and running modern AI models. The US imposed export restrictions on such chips in 2022, but following a meeting between Donald Trump and Chinese President Xi Jinping last year, the US president agreed to permit their sale to approved Chinese customers.
In January, Washington relaxed licensing rules for the H200, allowing export applications to be reviewed on a case-by-case basis with security requirements in mind. This followed months of lobbying by Jensen Huang.
The chips are manufactured in Taiwan by TSMC. The agreement also provides for a 25% tariff on eligible accelerators imported into the US before being shipped to China. Washington approved shipments to ten Chinese companies, including Alibaba, Tencent, and ByteDance.
However, Beijing restricted H200 imports in an effort to reduce dependence on American technology. According to Reuters, around 600,000 accelerators are awaiting clearance to enter mainland China and are stored in Hong Kong warehouses. Meanwhile, several major Chinese internet companies have received roughly 10,000 such chips each in recent months.
