Nvidia’s H200 artificial intelligence accelerators have finally arrived at Chinese technology giants like ByteDance and Tencent, which have each received approximately 10,000 units. Despite Washington approving export licenses valued at $10 billion, actual shipments remain tightly constrained by manufacturing bottlenecks at TSMC and shifting regulatory stances from Beijing.
The Reality of a $10 Billion Licensing Gap
For months, the export machinery moved significantly faster than physical silicon. Back in July 2026, Jeffrey Kessler, the Under Secretary of Commerce for Industry and Security, testified before the House Foreign Affairs Committee that the volume of delivered H200 chips was effectively negligible. Out of roughly two million chips ordered by Chinese firms for 2026—spanning approximately ten approved buyers including Alibaba, Tencent, ByteDance, and JD.com—Nvidia’s total global inventory stood at just 700,000 units.
This massive gap exposed a severe manufacturing constraint. The H200 relies heavily on TSMC’s 4-nanometer process node alongside advanced Chip-on-Wafer-on-Substrate (CoWoS) packaging. Because those exact packaging lines service Nvidia’s entire Hopper and Blackwell lineups, physical shipments lagged far behind paper approvals. That supply crunch forced emergency production talks to restart older Hopper-generation manufacturing lines.
Beijing’s Strategic Turnaround and Self-Sufficiency Drives
While Washington opened the door by clearing conditional export licenses starting in December 2025, Beijing enforced its own internal roadblocks. China initially restricted local firms from purchasing the hardware to protect its domestic semiconductor self-sufficiency drive. However, as the computing-power gap threatened frontier AI model development, the stance shifted into a targeted middle-ground solution.
According to reports from the South China Morning Post, the Chinese government planned to let selected operators like Alibaba Group Holding buy limited numbers of the H200 to temporarily ease training bottlenecks. Companies including ByteDance and start-up DeepSeek were informed of upcoming approvals, though regulatory hurdles remain. Chinese officials have mandated that domestic companies must justify why they cannot use local processors—such as Huawei’s Ascend line—for specific workloads.
Regulatory Scrutiny and Enforcement Failures
The Bureau of Industry and Security continues to review H200 applications on a strict case-by-case basis. Buyers must certify that the hardware will not be deployed in military applications, nuclear development, or biological and chemical weapons programs. Yet, enforcement remains a persistent challenge for regulators.
During congressional oversight hearings, lawmakers raised alarms over separate enforcement failures. Evidence indicated that Nvidia’s advanced Blackwell architecture—specifically the GB200 NVL72 platform—may have leaked to Chinese entities for nearly a year through a regulatory loophole that the Commerce Department finally closed in May.
Concurrently, the Bureau of Industry and Security expanded its approved buyer lists to include entities like ZTE Kangxun Telecom, a unit of ZTE Corp. This addition drew sharp bipartisan scrutiny given ZTE’s previous designation as a national security threat under the 2018 National Defense Authorization Act. As ByteDance and Tencent integrate their initial 10,000-unit allocations of H200 accelerators, the friction between geopolitical export controls and raw enterprise compute demand continues to shape the global artificial intelligence landscape.
