Asia-Pacific weekend brief, HBM supply risk and China’s homegrown AI chips dominate attention

Asia-Pacific tech investors head into the weekend focused on two pressure points, high bandwidth memory supply from Korea and China’s shift to domestic AI accelerators. The first constrains model training roadmaps across the United States and Europe, the second reshapes how global firms source compute in China in light of export controls.

The supply picture matters because SK Hynix and Samsung remain the primary HBM suppliers alongside Micron, and advanced packaging throughput at TSMC and Samsung Foundry continues to gate AI server output. Both Korean firms have previously guided to tight HBM availability through 2026 as they convert DRAM lines and qualify HBM3E and early HBM4, a view echoed by hyperscalers that cited packaging as a bottleneck after Computex.

HBM pricing and allocation remain closely tied to long term agreements with Nvidia, AMD, and custom AI ASIC buyers, according to recent earnings calls from SK Hynix chief executive Kwak Noh Jung and Samsung memory head Lee Jung bae. Micron chief executive Sanjay Mehrotra has also signaled a higher mix of HBM in DRAM revenue as the firm ramps volumes, which keeps pressure on general purpose DRAM availability for the rest of the year.

In China, large model developers including Baidu, Alibaba Cloud, and ByteDance are expanding training and inference pilots on domestic accelerators such as Huawei’s Ascend as access to high end Nvidia parts remains restricted. Alibaba’s T-Head unit continues to push ARM based Yitian server chips in its cloud, and Huawei’s computing business has leaned on the Ascend 910B with local ODMs to assemble AI racks that meet government procurement and data localization requirements.

Chinese regulators, led by the Cyberspace Administration of China and the Ministry of Industry and Information Technology, have kept model filings and compute traceability on the agenda, which nudges cloud customers toward infrastructure that can demonstrate onshore control. That influences procurement plans for multinationals that run mainland workloads through local partners or joint ventures.

For global technology teams planning capacity and compliance, the near term takeaway is practical. AI server lead times still hinge on HBM and packaging slots in Korea, Taiwan, and Japan, and China facing workloads will increasingly encounter Huawei aligned stacks inside regulated clouds. Mapping suppliers and alternative configurations now lowers the risk of missed deployment targets in the second half.