China Fast-Tracks Domestic AI Chips on Government List Ahead of US Export Shift

In a significant pivot toward technological self-reliance, China's Ministry of Industry and Information Technology (MIIT) has officially added domestically produced AI chips to its "Safe and Reliable" government procurement list for the first time. This strategic update, executed on November 15, 2025, predates President-elect Trump’s recent move to conditionally ease export restrictions on Nvidia by approximately three weeks.

The timing and scale of this addition signal a clear intent from Beijing: regardless of shifting winds in Washington, the mandate for state-owned enterprises is now explicitly focused on domestic hardware.

A Historic First for Chinese Silicon

For years, China's government procurement guidance was dominated by foreign suppliers. As recently as 2024, reports indicated that roughly 80% of the list was comprised of foreign technology, primarily from giants like Nvidia and AMD. The November update fundamentally alters this landscape, shifting the balance to a near 50/50 split between domestic and foreign options.

The inclusion of 15 specific AI chip models from local heavyweights marks a critical milestone. Among the most notable additions are Huawei’s HiSilicon Ascend series and Cambricon Technologies' MLU series. According to MIIT’s November 15 release, these chips are now officially designated for use in state-owned enterprises and government projects.

This isn't a token gesture. The China Electronics Standardization Institute projects that these specific models are capable of covering over 40% of the government’s AI procurement needs for 2026. The move aligns with a broader surge in domestic capability; IDC Research estimated on December 9 that domestic chips now account for 30% of China’s AI chip market, a substantial leap from just 10% in 2024.

MIIT characterized the decision as a "critical step in building a secure, independent supply chain," emphasizing technological sovereignty. While recent geopolitical developments have garnered headlines, Chinese officials reiterated on December 10 that this domestic push is "not influenced by external policies" but rather driven by long-term national security goals.

Preempting the Trump-Nvidia Export Deal

The timeline reveals a proactive strategy rather than a reactive one. China’s list update occurred weeks before news broke on December 8 regarding President-elect Trump’s plan to allow Nvidia to export its H200 AI chips to China.

While the incoming U.S. administration’s policy offers a potential thaw—permitting Nvidia exports in exchange for a 25% revenue share to the U.S. Treasury—China had already cemented its domestic preference. This creates a complex environment for U.S. tech firms. While Nvidia CEO Jensen Huang welcomed the "policy clarity" and the opportunity to responsibly export H200 chips, the market they are re-entering is rapidly changing.

Analysts suggest that even with the U.S. door cracking open, China is unlikely to abandon its defensive posture. The MIIT’s directive mandates 60% domestic sourcing in government contracts for key AI hubs like Beijing and Shanghai. Furthermore, China has ramped up subsidies for 2025 to $15 billion, up from $10 billion the previous year, specifically to support the ecosystem surrounding these newly listed chips.

Technical Viability and Market Realities

The question remains: can these domestic chips compete? The answer depends on the specific use case.

The newly listed Huawei Ascend 910B is touted as having 30% better energy efficiency than its 2024 predecessors, with built-in encryption that complies specifically with Chinese national standards—a key differentiator for government clients concerned with data security. Similarly, Cambricon’s MLU290 claims to offer 20% faster inference speeds compared to Nvidia's restricted A100 chips for edge computing tasks.

However, a performance gap persists at the high end. While capable, domestic options generally trail the raw training speed of Nvidia’s premier offerings. The approved export model, the Nvidia H200, still offers roughly 2x faster training performance than current Chinese equivalents.

This discrepancy points toward what Bloomberg described on December 9 as a potential "hybrid procurement" model. China may rely on domestic silicon for government data security and standard workloads—meeting the new "Safe and Reliable" criteria—while potentially utilizing permitted Nvidia imports for raw commercial performance where technically superior hardware is essential.

Ultimately, the November 15 list update serves as a signal that while China may still buy American chips, it is no longer willing to depend on them.