China Adds Domestic AI Chips to Official Procurement List for First Time
U.S. export restrictions have forced China to accelerate its push for technological self-reliance, spotlighting vulnerabilities in global AI supply chains. Beijing’s latest move integrates homegrown processors into government purchasing guidelines, potentially unlocking billions in sales for local firms. This policy shift challenges Nvidia’s dominance while exposing gaps in domestic chip performance.
The Ministry of Industry and Information Technology added AI chips from Huawei and Cambricon to the Information Technology Innovation List, known as Xinchuang. This catalog directs procurement for government agencies, public institutions, and state-owned enterprises, prioritizing domestic alternatives to foreign hardware. The inclusion marks the first time AI-specific processors have joined the list, following earlier additions of microprocessors replacing AMD and Intel equivalents, and operating systems supplanting Microsoft Windows.
Procuring entities now face incentives to select these chips for data centers and AI workloads, with tech giants like Alibaba and Tencent expected to align purchases. The policy responds to U.S. controls tightened since 2022, which barred advanced Nvidia GPUs like the H200 from export to China. Trump’s Monday announcement to lift some restrictions could ease immediate pressures, but Beijing’s strategy emphasizes long-term independence.
Domestic chips lag in efficiency, consuming more electricity than Nvidia counterparts and requiring code rewrites for compatibility. A state-owned financial institute allocated 100 million renminbi ($14 million) for these processors this year, yet executives report idle hardware due to adaptation challenges. Recent subsidies for data centers, increased amid the push, can halve energy bills for compliant facilities.
Cambricon’s MLU series and Huawei’s Ascend lineup target large language model training and inference, though benchmarks show 20-30 percent lower throughput on standard tests. The Xinchuang framework, expanded over the past three years, now encompasses servers, storage, and networking gear, fostering an ecosystem valued at over 1 trillion renminbi annually. Vendors must certify products through rigorous audits, ensuring interoperability with state-approved software stacks.
This procurement pivot aligns with China’s 14th Five-Year Plan, aiming for 70 percent domestic IT sourcing by 2025. Policymakers view short-term inefficiencies as necessary for sovereignty. One official remarked, “The growing pains are unavoidable… But we have to get there,” underscoring resolve amid U.S.-China tech frictions.
For U.S. audiences, the development raises stakes in bilateral trade talks, as American firms like Nvidia eye resumed sales while competitors gain ground. Beijing’s subsidies, totaling 200 billion renminbi since 2023, bolster chipmakers’ R&D, narrowing the performance gap projected to close by 2027. Enterprises adapting to Xinchuang report 15 percent cost savings on compliant builds, offsetting hardware premiums.
The list’s expansion signals broader decoupling, with implications for global semiconductor markets projected to reach $1 trillion by 2030. Huawei, facing entity list sanctions since 2019, leverages the policy to scale Ascend 910B production to 1 million units yearly. Cambricon, post-IPO in 2020, eyes 50 percent revenue growth from public sector deals.
Challenges persist in scaling fabs for 7nm and below nodes, where SMIC trails TSMC by two generations. U.S. stakeholders monitor for spillover effects on allied supply chains, as Japan and South Korea firms adjust strategies. This move reinforces AI as a national security priority, mirroring Washington’s CHIPS Act investments exceeding $50 billion.
Overall, China’s procurement mandate accelerates a parallel AI ecosystem, compelling multinationals to localize operations or risk exclusion. With domestic market share for AI chips rising from 10 percent in 2023 to 25 percent this year, the policy cements Beijing’s bet on innovation under constraints.
