The strategic landscape of the global artificial intelligence race has become a paradox of policy, as China reportedly moves to slow imports of advanced AI chips from Nvidia, immediately following a conditional loosening of export rules by the United States. This unusual dynamic reveals a calculated pivot by Beijing, which is leveraging a slight concession from Washington as a strategic opportunity to accelerate its long-term goal of technological self-sufficiency. Rather than rushing to purchase newly-available American hardware, China is signalling its intent to protect its nascent domestic chip industry, even at the risk of temporarily slowing its AI development pace.

The friction is centered on Nvidia’s most advanced graphics processing unit approved for sale to China, the Hopper-generation H200 chip. The recent decision by the US, announced by former President Donald Trump, partially reversed a previous ban on high-end chips and permitted the export of the H200 to approved Chinese customers. This approval was not unconditional, however, reportedly tied to a unique 25% revenue-sharing tax payable to the US government, an aggressive mechanism intended to both secure an economic benefit and maintain leverage. The White House framed the move as a pragmatic compromise to keep American firms competitive in a vital market while denying Beijing access to the cutting-edge Blackwell series, such as the B300 chip, which remains restricted. China’s response has been to instruct domestic technology companies to temporarily halt H200 purchases, a pause intended to prevent premature stockpiling before the government finalizes its own import policies.[1][2][3]

At the core of Beijing’s deliberation is a proposed quota system, a potential non-tariff barrier that would fundamentally restructure the Chinese AI hardware market. This regulatory mechanism is being considered to require domestic buyers of imported chips, such as the Nvidia H200, to also commit to purchasing a set percentage of domestically produced AI accelerators. Such a quota would guarantee market share for homegrown alternatives and force a shift in the supply chain, transforming American-made chips from a necessary component into a heavily-regulated, expensive add-on. This strategy is an escalation of previous, less formalized guidance that had already seen the Cyberspace Administration of China direct major tech firms like Alibaba and ByteDance to cancel or pause orders for earlier China-specific Nvidia chips, the H20 and RTX Pro 6000D, following Beijing’s assessment that local alternatives were now capable of matching or surpassing their performance.[4][5][6][1]

This protective stance is underpinned by a massive, state-directed industrial policy and a rapidly maturing domestic ecosystem. China's National Integrated Circuit Industry Investment Fund, commonly known as the Big Fund, established its third phase with a registered capital of 344 billion yuan (approximately $47.5 billion), a significant increase from previous phases. This fund is explicitly targeting investments across the entire integrated circuit supply chain, with a strong focus on high-value-added components like high-bandwidth memory and AI semiconductors, reinforcing the long-term drive for "self-sovereignty" in the face of ongoing geopolitical trade volatility.[7][8] The goal is not just to replace foreign hardware but to build a fully self-reliant AI technology stack. State-funded data center projects have been mandated to exclusively use domestically manufactured AI chips, and homegrown accelerators from players like Huawei and Cambricon are now included on government-approved supplier lists for the first time.[9][10]

Huawei's Ascend series has emerged as the most viable homegrown challenger to Nvidia's dominance. While the Ascend 910C chip is generally considered to be less powerful than the newly approved Nvidia H200, it is reportedly on par with Nvidia’s restricted H100 chip in some metrics, and a new version, the Ascend 910D, is slated to further narrow the performance gap.[11][12][13] Crucially, Chinese companies are optimizing their algorithms for these domestic chips, even adopting new standards like the FP8 data format advocated by a leading domestic AI startup, to squeeze more efficiency from the available hardware. Furthermore, the focus is shifting from per-chip performance to cluster-level performance, with domestic firms emphasizing dense clustering and high-speed interconnects to offset the technical lag. Chinese foundries, including SMIC, are planning to double their output of 7-nanometer processors, which are used for Huawei’s Ascend chips, with reports indicating China aims to triple its overall AI chip production in the coming year. This rapid scaling, supported by the Big Fund's enormous capital injection and the construction of new fabrication facilities, suggests Beijing views its domestic industry as mature enough to withstand a period of controlled, limited foreign competition.[14][4]

The global implications of China’s protectionist move are profound, signaling a deepening bifurcation in the world’s AI industry. The US policy to selectively loosen export controls in a bid to maintain a foothold and influence on China’s tech stack appears to be backfiring, as Beijing seizes the moment to institutionalize its preference for local vendors. By deliberately imposing new hurdles for even the conditionally approved Nvidia chips, China is effectively accelerating the decoupling process and pushing its major tech firms to fully commit to domestic hardware. This guarantees that China’s domestic AI market will develop a separate software and hardware ecosystem, which could complicate future interoperability and market access for foreign firms. The move ensures that the next wave of advanced AI development will proceed along two distinct, competitive paths: the US-led Western stack built on Nvidia's most powerful, unrestricted chips, and an increasingly capable, state-backed Chinese stack built on Huawei’s Ascend and other domestic alternatives.