China's AI Chip Budget Share Climbs to 46% as HBM Shortage Limits Growth

Chinese artificial‑intelligence labs are increasingly designing custom processors to pair tightly with their models, aiming for better efficiency and lower inference costs. Start‑ups such as DeepSeek and Zhipu AI have begun collaborations with domestic chip designers, while major cloud providers and hardware firms are investing in in‑house silicon projects.

Bloomberg Industry Research finds that, over the next 12 months, Chinese firms will allocate 46% of AI‑chip spending to domestically produced chips, up from about 30% today. The shift is driven by U.S. export restrictions that cut off Nvidia H200 supplies, prompting cloud giants like Tencent, Alibaba and Huawei to adopt home‑grown solutions such as Huawei Ascend, HaiGuang and Cambricon. However, experts warn that the lack of high‑bandwidth memory (HBM) is the real bottleneck. China’s demand for AI chips is estimated at 4.2 million units this year, but only about 2.6 million can be supplied, leaving a gap of roughly 1.6 million HBM parts. The HBM market remains dominated by SK Hynix, Samsung and Micron, and domestic producer CXMT can currently produce only about 5,000 HBM chips per month, with larger volumes not expected until 2026‑2027. Government policy aims to spend about 2 trillion RMB on data‑center construction and to secure at least 80% of core chip supply domestically, underscoring the strategic importance of closing the HBM gap.