Chinese artificial intelligence firms are ramping up purchases of Nvidia’s H20 chips as access to high-performance AI hardware becomes increasingly restricted due to U.S. export bans.
With Nvidia’s most powerful chips—A100, H100, A800, and H800—now off-limits, China’s AI sector has turned to the H20 as one of the last available options.
The surge in demand is being driven by the widespread adoption of DeepSeek AI, an emerging Chinese alternative to OpenAI, which requires substantial computing resources.
With the U.S. imposing stricter controls on semiconductors, memory chips, and AI development tools, Chinese firms are scrambling to secure what remains available before further restrictions take effect.
How U.S. AI Chip Bans Reshaped Nvidia’s China Strategy
Washington first banned Nvidia’s A100 and H100 GPUs in October 2022, citing national security concerns. Nvidia attempted to comply by releasing the A800 and H800, downgraded versions designed to stay within U.S. regulatory thresholds. However, these too were blacklisted in late 2023, forcing the company to develop the H20, a further downgraded model meant to remain exportable under current restrictions.
Despite this adaptation, the H20’s future remains uncertain. As reported earlier today, U.S. policymakers are now reviewing additional restrictions that could place even lower-performance AI chips under export controls. If implemented, these new rules would cut off Nvidia’s last AI hardware supply to China, forcing companies to rely entirely on domestic alternatives.
DeepSeek AI’s Growth Strains China’s AI Infrastructure
DeepSeek AI has quickly established itself as a major player in China’s AI ecosystem, providing businesses with a local alternative to Western AI models. Its widespread adoption has led to an increased demand for Nvidia’s H20, as companies require substantial computational power to train and deploy DeepSeek-powered applications.
However, the strain on China’s AI infrastructure is becoming apparent. Due to overwhelming demand, DeepSeek earlier this month temporarily suspended API refills in an effort to stabilize server operations.
U.S. Expands Semiconductor Restrictions, Targeting AI Supply Chains
The U.S. has not only blocked advanced AI chips but is also tightening restrictions on other critical semiconductor components. In December 2024, Washington imposed a ban on high-bandwidth memory (HBM) chip exports, preventing companies like Samsung and SK Hynix from supplying high-performance memory modules to Chinese AI firms.
Additionally, the Biden administration is reviewing measures to further restrict ChangXin Memory Technologies (CXMT) and Semiconductor Manufacturing International Corp. (SMIC), two of China’s largest semiconductor manufacturers. Expanding the U.S. trade blacklist to include these firms would severely limit China’s ability to produce AI-optimized chips domestically.
Washington has also increased pressure on allied countries. The U.S. has successfully pushed the Netherlands and Japan to restrict the servicing and maintenance of lithography machines in China, a move aimed at slowing China’s ability to fabricate advanced AI processors.
China’s Response: Developing Domestic Semiconductor Alternatives
Faced with mounting restrictions, China is rapidly investing in semiconductor self-sufficiency. While extreme ultraviolet (EUV) lithography remains restricted due to export bans, Chinese firms are advancing in deep ultraviolet (DUV) lithography, allowing them to continue producing chips at slightly older process nodes.
Chinese tech giant Huawei has already demonstrated progress in bypassing U.S. restrictions. The company’s Kirin 9000S processor, reportedly fabricated using a 7nm process by SMIC, has raised concerns in Washington that China may be advancing its AI chip production capabilities faster than expected.
China is also looking to indirect AI chip supply chains. Despite U.S. bans, reports suggest that Chinese firms have been acquiring restricted AI chips through intermediaries in Southeast Asia, a loophole U.S. officials are now attempting to close.
Tech Giants Navigate the U.S.-China AI Divide
With geopolitical tensions escalating, major technology firms are taking different approaches to engaging with Chinese AI developers. Some companies are actively distancing themselves, while others continue to integrate Chinese-developed AI models despite U.S. scrutiny.
SoftBank recently froze its use of DeepSeek AI, shifting its focus toward expanding collaborations with OpenAI. This move reflects growing concerns about China’s AI sector and how it aligns with Beijing’s long-term strategic goals for artificial intelligence dominance.
Meanwhile, Microsoft has added DeepSeek’s openly available R1 model to Azure AI Foundry platform, providing developers access to the model despite ongoing investigations into whether DeepSeek AI improperly accessed OpenAI’s API data. Similarly, GitHub has integrated DeepSeek R1 into its AI offerings, raising concerns over the model’s origins and whether it benefits from U.S. intellectual property.
U.S. Considers DeepSeek AI Ban Over National Security Risks
The widespread adoption of DeepSeek AI has triggered alarms in Washington. Congress is debating a potential ban on DeepSeek AI, with lawmakers arguing that its technology could be used for state surveillance, propaganda, and military applications.
Texas has already taken independent action, enforcing a statewide ban on DeepSeek AI and other Chinese tech applications. This follows previous bans on TikTok and Huawei, highlighting ongoing U.S. concerns over the security risks posed by Chinese AI models.
Meanwhile, OpenAI has reinforced its API security policies and is conducting an internal investigation into whether DeepSeek AI’s developers misused its API for unauthorized model training.
Nvidia Warns of Economic Risks from Overreaching Sanctions
While the U.S. government tightens semiconductor restrictions, Nvidia has cautioned against excessive regulations that could backfire in the long run. The company, which remains a dominant AI chip supplier, has raised concerns that aggressive U.S. trade policies could accelerate China’s efforts to develop a self-sufficient semiconductor industry.
Nvidia continues to generate substantial revenue from its AI chip business in China, even under restrictions. However, if the H20 is added to the U.S. export blacklist, it would eliminate Nvidia’s last remaining AI chip sales in China, forcing the company to restructure its business strategy.
At the same time, China’s domestic chip industry is making rapid progress. If Beijing successfully develops competitive AI chips, U.S. restrictions could lose their effectiveness, shifting the balance of technological leadership in AI hardware.
The Global AI Industry Faces Fragmentation
Washington’s semiconductor restrictions represent a long-term strategy to contain China’s AI development. However, China’s continued investments in alternative semiconductor technology indicate that these restrictions may only be delaying an inevitable shift.
It appears that a bifurcated global ecosystem is emerging. The U.S. and its allies dominate high-end AI hardware and cloud-based AI services, while China is focusing on domestic chip production and closed-loop AI models that do not rely on Western technology.
The coming years will determine whether China can overcome semiconductor limitations quickly enough to remain competitive. If its AI industry manages to advance despite U.S. sanctions, the global AI race could shift in unpredictable ways.
Table: AI Model Benchmarks – LLM Leaderboard
[table “18” not found /]Last Updated on March 3, 2025 11:28 am CET
