AI and Geopolitics: Nvidia Adjusts Its Strategy in Response to U.S. Restrictions
Since 2022, the U.S. government has imposed restrictions on the export to China of advanced chips used to train artificial intelligence models. These measures were intended to limit Beijing’s access to critical technologies capable of supporting its military and industrial ambitions1.
Nvidia, the global leader in AI processors, was directly affected by these sanctions, which banned the sale of its flagship A100 and H100 models in China. In response, the company suspended certain shipments while preparing a technological counteroffensive: the development of alternative AI chips specifically designed to comply with the thresholds imposed by Washington.
In 2025, Nvidia is set to release these “modified” models, the launch of which would mark a partial resumption of its operations in China.
Custom AI chips designed to circumvent limitations without violating the embargo
The new chips currently in development are reportedly codenamed H20, L20, and L2. These are variants of the Hopper and Ampere architectures, but certain specifications (interconnect speed, memory bandwidth, FP16/FP32 computing capabilities) have been intentionally scaled back to remain below U.S. regulatory thresholds2.
This strategy allows Nvidia to offer its Chinese customers solutions that are always tailored to generative AI, while strictly complying with regulations. It illustrates the emergence of a geopolitics of controlled performance: selling without over-empowering.
Why does China remain a key market for artificial intelligence?
Despite the restrictions, China remains one of the world’s leading markets for AI chips:
- Explosive growth in cloud computing and data centers,
- A surge in AI startups and public projects,
- Growing need for infrastructure to train Chinese language models,
- Ambitious national plans to develop sovereign AI.
Tech giants such as Baidu, Tencent, Alibaba, and ByteDance have expressed interest in these alternative chips in order to maintain their ability to innovate despite sanctions3.
A delicate balance between innovation, dependence, and regulation
For Nvidia, the challenge is twofold: maintaining its market share in China without triggering a new wave of U.S. restrictions. Every technical adjustment is closely scrutinized by the authorities. Even the slightest software or hardware optimization could be interpreted as a strategic workaround.
At the same time, China is developing its own alternatives: Huawei (Ascend 910B), Cambricon, Loongson… but these chips still lag behind in terms of maturity, software ecosystem, and optimization for LLM models.
The global ecosystem is thus caught in a delicate balancing act between technological competition, digital sovereignty, and geopolitical pressures.
What are the challenges facing the global artificial intelligence ecosystem?
This situation has several systemic effects:
- Technological fragmentation: architectures that vary by geographic region,
- Disruptions in supply chains (TSMC, ASML, Samsung),
- Potential slowdown in global innovation in the event of a decoupling of standards,
- Accelerating AI autonomy in countries subject to restrictions.
Against this backdrop, the global governance of AI is being reshaped.
Between industrial strategy and geopolitical calculations: what comes next?
Nvidia finds itself in a unique position: it is both a technology company and a geopolitical actor. Its decisions influence not only access to AI, but also the balance of power among economic blocs.
It remains to be seen whether this moderate recovery in China will be tolerated by U.S. regulators in the long term. Or whether, on the contrary, it will lead to a stricter review of the export framework, reinforcing the trend toward technological decoupling between the United States and China4.
References
1. U.S. Department of Commerce. (2022). Export controls on advanced computing and semiconductor manufacturing items to China.
https://www.commerce.gov/
2. Reuters. (2025). Nvidia prepares new AI chips for China to comply with U.S. rules.
https://www.reuters.com/
3. South China Morning Post. (2025). Chinese tech giants eye Nvidia’s H20 chips amid export restrictions.
https://www.scmp.com//
4. Brookings Institution. (2024). The Geopolitics of Semiconductor Supply Chains and AI Decoupling.
https://www.brookings.edu/
