Nvidia Corp. has delivered approximately 10,000 H200 AI chips each to Chinese tech giants ByteDance Ltd. and Tencent Holdings Ltd. [1], [2].
This movement of hardware marks the first flow of these specific high-end chips into mainland China following a period of strict U.S. export restrictions. The delivery signals a shift in how the U.S. manages the balance between national security concerns and the commercial interests of American semiconductor firms.
U.S. authorities granted limited export approvals for the H200 chips in 2023 [4]. These approvals were later complemented by Chinese regulators, who allowed top domestic AI firms to purchase the hardware [3], [5].
Reports indicate that the U.S. cleared roughly 10 Chinese companies to purchase the H200 chips [3]. While the broader trade environment remains tense, these specific shipments represent a targeted easing of restrictions for a small group of vetted entities.
ByteDance and Tencent each received 10,000 units [1], [2]. The chips are destined for the companies' Chinese subsidiaries to bolster their artificial intelligence capabilities. Additional shipments are expected to follow as these companies integrate the hardware into their existing infrastructure [2].
The H200 is a critical component for training large-scale AI models. By securing these chips, ByteDance and Tencent can maintain their competitive edge in the global AI race, despite the ongoing geopolitical friction between Washington and Beijing.
“Nvidia has delivered approximately 10,000 H200 AI chips each to ByteDance Ltd. and Tencent Holdings Ltd.”
The delivery of H200 chips suggests that the U.S. government is adopting a surgical approach to export controls rather than a blanket ban. By allowing a small number of cleared companies to access high-end silicon, the U.S. maintains oversight of who possesses the technology while preventing Nvidia from losing its most significant foreign market entirely. For Chinese firms, this access is vital to avoid a complete reliance on domestic chip alternatives, which may not yet match the performance of Nvidia's architecture.
