Beijing has taken its most aggressive step yet to decouple its critical technology infrastructure from Western suppliers. China has issued new internal guidance requiring all state-funded artificial intelligence data center projects to utilize exclusively domestically manufactured chips. The mandate effectively bans the procurement of processors from foreign industry leaders, including Nvidia, AMD, and Intel, for any government-backed computing facility.
The directive, which has not been publicly announced but is being actively enforced across provincial and municipal projects, represents a significant escalation in China’s drive for technological self-sufficiency. While previous government efforts relied heavily on subsidies and preferential treatment to encourage the adoption of local hardware, this new guidance establishes a hard regulatory boundary for state-affiliated infrastructure investments.
The immediate beneficiaries of this policy shift are China’s domestic AI chipmakers, primarily Huawei Technologies, whose Ascend series of processors are currently the only viable domestic alternative capable of handling large-scale AI training workloads. Other domestic players, such as Cambricon and Moore Threads, are also expected to see significant order increases as local governments scramble to comply with the new procurement rules.
Accelerating the Self-Sufficiency Drive
The mandate arrives at a moment when China is rapidly expanding its national computing infrastructure. The government has launched a massive initiative to build a nationwide network of AI data centers, designed to provide the foundational computing power required to support the country’s booming generative AI sector. By ensuring that this infrastructure is built entirely on domestic silicon, Beijing is attempting to insulate its AI ecosystem from the persistent threat of tightening US export controls.
The scale of this infrastructure build-out is massive. The government has committed to a two trillion yuan ($295 billion) data center investment over the next five years, with local suppliers including Huawei required to provide at least 80 percent of the technology. China’s domestic chipmakers have already seized 41 percent of the local AI market, and that share is expected to grow rapidly as the mandate takes effect. Morgan Stanley estimates that China’s AI chip self-sufficiency level will reach 70 percent by the end of this decade, up from 42 percent in 2025.
The policy shift is expected to dramatically alter the market dynamics for AI hardware in China. A Bloomberg Intelligence survey of Chinese technology executives conducted in June found that Chinese companies expect to spend 46 percent of their AI accelerator budget in the next 12 months on locally made chips, up from 30 percent currently. This rapid shift is being driven primarily by the mandatory adoption of domestic processors by state-owned enterprises and government-funded computing hubs, with Huawei, Cambricon, and Hygon among the primary beneficiaries.
The Challenge of Implementation
While the mandate provides a massive guaranteed market for domestic chipmakers, its implementation presents significant engineering and financial challenges for the data center operators. Transitioning away from Nvidia’s deeply entrenched hardware and software ecosystem is not a simple procurement swap. Nvidia’s Compute Unified Device Architecture (CUDA) platform has been the global industry standard for AI development for nearly two decades, and the vast majority of existing AI models and training pipelines are optimized specifically for it.
Forcing state-funded data centers to adopt Huawei’s Ascend chips or other domestic alternatives requires developers to rewrite and optimize large amounts of code to function on new, often less mature, software stacks like Huawei’s Compute Architecture for Neural Networks (CANN). Industry sources estimate that migrating existing workflows to domestic chips can add at least 50 percent in time and engineering costs.
Furthermore, while domestic chips are closing the performance gap in inference tasks (running trained models), they generally still lag behind Nvidia’s latest offerings in training efficiency (building new models from scratch). State-funded research institutes and universities, which rely heavily on these data centers, may face near-term productivity bottlenecks as they adapt to the mandated hardware.
Closing the Overseas Loophole
The new domestic chip mandate is part of a broader, comprehensive strategy by Beijing to secure its AI supply chain. However, the effectiveness of this strategy is complicated by the global nature of cloud computing. Currently, a significant loophole exists that allows Chinese AI firms to access advanced Nvidia chips through overseas data centers, bypassing both US export controls and Chinese domestic mandates.
While the new state-funding mandate secures the physical infrastructure within China’s borders, private Chinese tech giants continue to leverage offshore data centers to access restricted Nvidia hardware for their most demanding training workloads. This gap between the government’s self-sufficiency mandate and the private sector’s engineering preferences highlights the ongoing tension at the heart of China’s AI strategy.
As the domestic chip mandate takes effect, the focus will shift to how quickly companies like Huawei can scale production to meet the sudden surge in guaranteed demand, and whether their software ecosystems can mature fast enough to prevent the mandated transition from crippling the productivity of China’s state-backed AI research initiatives.