China´s artificial intelligence chip ambitions limited by HBM memory supply

A SemiAnalysis report finds high-bandwidth memory shortages, not foundry capacity, are the main bottleneck for China´s artificial intelligence chip scaling, constraining production despite stockpiles and foundry access.

A SemiAnalysis report argues that high-bandwidth memory shortages are the primary constraint on China´s artificial intelligence semiconductor buildout, outweighing manufacturing limits. Domestic foundries such as SMIC can produce sufficient processors, and companies like Huawei have foundry capacity tied to pre-cutoff stockpiling and external partners. Huawei´s Ascend 910C is cited as an example: the company has foundry capacity to manufacture 805,000 units annually through TSMC and SMIC, but that output cannot be realized because of insufficient HBM supply. Chinese firms accumulated roughly 11.4 million HBM stacks from Samsung before export controls tightened, yet those reserves are not enough to sustain large-scale artificial intelligence growth over the long term.

The report highlights the domestic path to HBM independence through memory manufacturers such as CXMT and the recent collaboration with YMTC. Converting standard DRAM production to HBM manufacturing requires specialized equipment, which is predominantly supplied by Western vendors, creating a secondary dependency. SemiAnalysis suggests China could reach competitive HBM3E production by 2026 if current investment trajectories continue and equipment restrictions remain stable. YMTC is reportedly preparing to enter DRAM production and could begin purchasing equipment by late 2025, leveraging its Xtacking hybrid bonding technology. That capability matters because HBM requires stacking at least 16 memory chips with high precision, a technique YMTC has applied in NAND products.

The memory crunch effectively blunts Chinese manufacturing capacity and preserves an advantage for Western competitors such as NVIDIA and AMD, despite Beijing´s substantial semiconductor investments. The projected 2026 timeline depends on successful technology transfer and uninterrupted access to critical manufacturing tools. If export controls expand to cover additional HBM-related equipment, SemiAnalysis warns the development window for domestic HBM production could lengthen considerably, prolonging the supply bottleneck and limiting China’s artificial intelligence hardware scaling.

65

Impact Score

Sneha Goenka’s ultra-fast sequencing cuts genetic diagnosis to hours

Sneha Goenka, MIT Technology Review’s 2025 Innovator of the Year and an assistant professor at Princeton, helped build a sequencing pipeline that reduces genetic diagnosis from weeks to hours. Her work pairs cloud computing architectures with real-time streaming analysis to accelerate clinical genomics.

Why basic science deserves our boldest investment

The transistor story shows how curiosity-driven basic science, supported by long-term funding, enabled the information age and today´s Artificial Intelligence technologies. Current federal and university funding pressures risk undermining the next wave of breakthroughs.

35 innovators under 35 for 2025

mit technology review presents its 35 innovators under 35 for 2025, profiling young scientists, inventors, and entrepreneurs tackling climate change, disease, and core scientific challenges.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.