China has finally addressed the reports that 600,000 advanced Nvidia chips are sitting in Hong Kong waiting for import approval. At the regular press briefing on November 9, foreign ministry spokesman Guo Jiakun said China's position on US chip exports to China is consistent, and that for specific matters reporters should ask the competent Chinese authorities, according to Reference News. The question itself leaned on a Reuters report, which put the chips in Hong Kong, bound for the mainland, with import approval still pending on the Chinese side.
An AFP journalist relayed the scoop. Who the buyers are, or which chip models exactly, nobody has disclosed. That said, the hunger of Chinese internet giants for Nvidia hardware is hardly a secret. Reuters reported on October 20 that Chinese internet companies have been buying Nvidia graphics cards in bulk — cards not built for AI data centers that nonetheless handle AI inference workloads — and that Nvidia has struggled to keep up with the surge in orders.
Approval may be stuck, but domestic alternatives keep moving. On October 20, DeepSeek open-sourced its in-house inference framework DeepInfer, listing Huawei's Ascend 950 NPUs and Nvidia's SM100 GPUs as equal build options from the same code tree, switchable with a single environment variable. The Information later reported that DeepSeek is working with Huawei to design its next-generation AI chip, with Huawei engineers embedded in the project and next-generation training chips to be prioritized.
The market noticed. On November 9, China's A-share computing-chip sector rallied. Jiemian-reported data cited by Financial World showed Cambricon leading gains of more than 8% intraday, while the STAR chip design index rose over 2%. GF Securities, per Sina Finance, argued the same day that US midterm elections have tightened constraints on compute buildout, and that the domestic compute chain is positioned to break through.
Whether the 600,000 chips clear customs, nobody knows yet. Either way, the episode suggests the most sensitive variable in the global compute supply chain has quietly moved — not through any single shipment, but from manufacturing capacity to approval.