Nvidia has announced plans to recommence sales of its H20 artificial intelligence (AI) chips to China, marking a significant development after a period of regulatory uncertainty. The company is currently filing applications with the U.S. government and anticipates receiving the necessary licenses shortly, enabling prompt delivery of these chips to Chinese clients. Additionally, Nvidia is introducing the RTX Pro, a new chip specifically designed for the Chinese market, which complies fully with existing regulations and is tailored for applications in digital manufacturing, such as smart factories and logistics.
The H20 chip has been central to the ongoing technological tensions between the United States and China. Although it is not Nvidia’s most advanced AI processor, the H20 represents the most powerful chip that the company can legally export to China under current U.S. export controls. This chip is optimized for inference tasks, which involve running pre-trained AI models for everyday applications, rather than for training new AI systems from the ground up.
Earlier this year, major Chinese technology firms, including ByteDance, Alibaba, and Tencent, proactively stockpiled H20 chips in anticipation of potential tightening of export restrictions. The H20’s appeal lies in its superior memory bandwidth compared to domestic Chinese alternatives and its seamless integration with Nvidia’s widely adopted software ecosystem, which facilitates easier deployment.
The regulatory landscape has been marked by significant fluctuations. In April, the Trump administration imposed restrictions on H20 sales, a move that threatened to cost Nvidia between $15 billion and $16 billion in revenue, based on the substantial purchases made by Chinese firms in the first quarter alone. These restrictions targeted chips exceeding specific performance thresholds, including a total memory bandwidth of 1,400 gigabytes per second and an input/output bandwidth of 1,100 GB per second.
However, these restrictions were short-lived. Following a high-profile meeting between Nvidia CEO Jensen Huang and President Trump at Mar-a-Lago, the administration reconsidered its stance. Reports suggest that the White House’s decision to reverse the ban was influenced by Nvidia’s commitment to invest heavily in U.S.-based infrastructure. Subsequently, Nvidia announced plans to build AI data centers in the United States, with investments potentially reaching up to $500 billion over the next four years, in collaboration with partners such as TSMC.
This series of policy reversals has drawn criticism from U.S. lawmakers who argue that such inconsistencies undermine efforts to limit China’s advancements in AI capabilities. The case of DeepSeek, a Chinese startup that developed an advanced AI model using Nvidia’s H800 chips—predecessors to the H20—illustrates these concerns. Despite the U.S. banning the sale of H800 chips to China in October 2023, Chinese suppliers have reportedly found ways to circumvent these restrictions.
In a statement, Nvidia spokesperson Hector Marinez highlighted that CEO Jensen Huang has been engaging with officials in both Washington and Beijing to emphasize the global benefits of AI for business and society.
This development underscores the delicate balance U.S. policymakers must maintain between national security interests and commercial considerations. Given the rapid evolution of the tech landscape in 2025, further policy shifts in this arena are likely.
