China’s CIIE spotlights domestic AI stacks as Korea’s HBM supply stays tight into 2026

China’s big clouds used Shanghai’s China International Import Expo this week to pitch enterprise AI built on local chips, while Korean memory suppliers kept attention on high bandwidth memory tightness that persists into 2026. The twin signals in Shanghai and Seoul matter for CIOs and component buyers setting next year’s AI infrastructure budgets.

CIIE runs through next week and is a major year end contracting venue for state owned enterprises and large private buyers. With United States controls still restricting advanced Nvidia shipments, Chinese vendors are concentrating offers on Huawei Ascend based racks and domestic accelerators for inference and model fine tuning.

Alibaba Cloud promoted Qwen based services bundled with on premises options, Baidu highlighted Ernie Enterprise on Baidu AI Cloud, and ByteDance’s Volcano Engine marketed Doubao APIs to industrial and finance customers. MIIT’s localization push under the Xinchuang program and procurement guidance from state asset regulators continue to favor domestic stacks, which gives Huawei’s Ascend 910B and 310B ecosystems practical momentum.

Price competition remains intense. Baidu and Alibaba cut API and model hosting prices in 2024 to win workloads, and sales teams this week leaned on total cost of ownership math that swaps Nvidia H100 class imports for Huawei clusters coupled with service credits. For foreign vendors with China exposure, the near term addressable market is moving further toward software and services layered on local hardware.

In Korea, HBM supply discipline is still the bottleneck that defines AI server delivery schedules and unit economics. SK Hynix leads shipments in HBM3 and HBM3E for Nvidia accelerators, Samsung has increased qualification wins and has guided for higher HBM output, and Micron remains a third source for select customers. Industry trackers such as TrendForce have kept HBM utilization and pricing firm through 2025, and purchasing managers say lead times for advanced stacks remain extended into the new year.

Seoul’s policy remains aligned with capacity expansion. The K Chips Act framework, which raised tax credits for strategic semiconductor investments, continues to support SK Hynix’s Cheongju and Yongin projects and Samsung’s Pyeongtaek lines, with packaging and substrate additions intended to ease eventual HBM4 transitions. That cadence limits near term relief for GPU integrators and cloud builders.

For global tech teams, the weekend takeaway is a widening stack divergence. China’s enterprise AI will run increasingly on Huawei and other domestic silicon with aggressive cloud pricing, while outside China the pace and cost of AI deployment still hinge on HBM constrained supply chains anchored in Korea.