Why are Nvidia chips being sold to China again? | Explained

The story so far: On December 8, U.S. President Donald Trump announced that Chinese firms will be able to import Nvidia’s H200 graphics processing units, provided the company pays the U.S. government a 25% revenue surcharge.

What are the chips used for?

Nvidia designs and develops graphics processing units, or GPUs. While GPUs are what drive digital displays and help with advanced workloads like video games, some are designed to help with AI development, such as by training large language models (LLMs), and performing other resource-intensive tasks. H200 is one generation behind the Blackwell architecture, which is the cutting edge of Nvidia’s product lineup. A key differentiator that Nvidia has is its proprietary CUDA software architecture, which it uses to improve the performance of its GPUs.

For geostrategic reasons, the U.S. has restricted or banned the export of highly advanced chips and the know-how to manufacture them since 2018. The U.S. has been joined in doing this by other countries whose firms hold critical leads in GPU and semiconductor technology, like South Korea, Japan and the Netherlands. One reason why the U.S. is wary of China having access to cutting edge technology is the dual-use implications. The U.S. wants to prevent China from arriving at technology breakthroughs first, especially in sensitive fields like defence. Maintaining a technological lead is also beneficial from a commercial point of view for U.S. firms, which would enjoy an advantage from having access to equipment and gear that their Chinese competitors don’t.

Why is export of H200 chips being allowed?

China has already faced a massive backlash to its semiconductor and AI industry from the U.S. Even previous generation H20 chips were subject to an export quantity quota, and reports indicate that the U.S. wanted visibility and control over who the end user for these technologies were. These restrictions spanned both the Joe Biden presidency and Mr. Trump’s terms. These limitations have driven China to invest enormous amounts of resources in research and development, fed by the country’s proceeds from its export-driven economy and generous support of industry champions like Huawei. The firm has been able to develop indigenous chipsets, and even an operating system for its new phones, that don’t rely on Android, which drives most smartphones other than Apple’s. Huawei’s indigenous phones use chips that use older processes, but boast decent performance. While tech restrictions arguably provide the U.S. with some breathing room to maintain its technological edge, firms like Nvidia have nevertheless sought to access the vast Chinese market. The firm successfully lobbied the White House to allow its previous generation chips to be sold in China.

Will China allow their purchase?

Nvidia and the U.S. government’s rationale is as follows: by allowing advanced GPUs to be sold to Chinese firms, the U.S. would take some momentum out of China’s efforts to develop and grow its own alternatives, such as those under development at Huawei. If Chinese firms are able to achieve their ends with H200 chips, the appetite for the more advanced B200 chips may not be pressing. The argument rests on the prospect that Nvidia can make money from Chinese sales, without disturbing the U.S.’s tech lead, while also not speeding up China’s research and development for foundational technologies. China knows of these strategies, and has in the past taken both paths before it: allowing firms to buy advanced chips from abroad, and in other cases mandating firms to use home-grown alternatives.

Reports indicate that the H200 chips that China will allow its firms to buy will be limited. That way, the country presumably hopes, the short-term needs of firms dependent on this technology will go on unimpeded, while a larger industry-level push for Chinese chip capabilities will reach U.S. levels. There remains a significant gap between the frontier of chip technologies and China’s answer to them.

China has been known to close technological gaps with far more limited access to technology in record time. Its DeepSeek LLM, developed by a relatively small firm at a lower cost than U.S. firms and with older chips, highlighted this to global policymakers. However, the U.S. and allied countries with a technological edge in different parts of the AI and chip ecosystem continue to see value in prolonging their pole position for as long as possible, perhaps until artificial general intelligence (AGI) is achieved.