How Chinese Companies Are Replacing Nvidia Chips With Domestic Alternatives at an Unprecedented Pace

U.S. Export Controls: The Primary External Catalyst

U.S.-China tensions are the primary external driver. Since the U.S. expanded export bans on advanced AI chips and placed firms like Z.ai (formerly Zhipu AI) on the Entity List in January 2025, Chinese companies have been effectively cut off from Nvidia's H100/H200-class hardware, forcing them to build domestic alternatives . In September 2025, China's Cyberspace Administration instructed major technology companies including Alibaba and Tencent to cease purchasing Nvidia chips, directing procurement toward domestic suppliers .

The $295 Billion State Mandate for Domestic Chips

China is preparing to spend approximately 2 trillion yuan ($295 billion) over the next five years on a nationwide network of AI data centers, with a mandate that at least 80% of core technology — including AI chips — be sourced from domestic suppliers such as Huawei . State-owned carriers China Mobile and China Telecom will operate these hubs . The plan, drafted by the National Development and Reform Commission (NDRC), constitutes the largest single procurement walk-away from U.S. chip suppliers in history . Some estimates suggest the total investment, when integrated with power grid upgrades, could reach 5 trillion yuan (~$735 billion) .

Proof of Performance: GLM-5 Trained on Huawei Ascend, Zero Nvidia

Zhipu AI (now Z.ai) trained its 744-billion-parameter GLM-5 model entirely on 100,000 Huawei Ascend 910C chips using the MindSpore framework — with zero Nvidia hardware . The model subsequently scored #1 on the Humanity's Last Exam (HLE) benchmark with 50.4%, surpassing Claude Opus 4.5, GPT-5.2, and Gemini 2.5 Pro . GLM-5.2, an updated open-weight model released in June 2026 under an MIT license, continued the trend . This achievement represents the first time a frontier-class AI model has been trained entirely without Nvidia hardware, demonstrating that Huawei's Ascend chip ecosystem is mature enough to compete with top-tier Western silicon .

Beneficiaries: Cambricon and Hygon Information Technology

Cambricon Technologies plans to more than triple its AI chip output in 2026, targeting 500,000 accelerators, including up to 300,000 units of its Siyuan 590 and 690 processors . The company reported its first-ever annual profit in 2025 (2.1 billion yuan / $306 million), compared with a 452 million-yuan loss in 2024, as revenue surged to 6.5 billion yuan from 1.2 billion yuan . Its Q1 2026 revenue hit $423 million . Hygon Information Technology crossed ¥40 billion in quarterly revenue, with its DCU series achieving "Day-0" compatibility with DeepSeek V4, matching Nvidia's software readiness . Both firms were among nine domestic AI chip vendors certified in May 2026 for China's "secure and reliable" government procurement list, unlocking institutional buyers .

Market Traction Moves Beyond Subsidies

Q1 2026 earnings reports showed Cambricon and Moore Threads with year-on-year revenue growth exceeding 150% — signaling that the transition from "subsidy-driven" to "product-driven" market traction is well underway . Chinese AI chip vendors also completed simultaneous (Day-0) adaptation of DeepSeek V4 in 2026, marking the transition from "lagging" to "simultaneous deployment" relative to Nvidia . Together, these data points confirm that the procurement shift is not a distant projection — it is already happening, reinforced at every level by policy, performance, and commercial reality.