Alibaba rises 4.8% as China may approve Nvidia H200 chip imports

Alibaba shares rose up to 4.8% in Hong Kong after reports that China may allow imports of Nvidia’s H200 chips as soon as this quarter.

Alibaba, which has invested heavily in AI infrastructure, is reportedly interested in ordering over 200,000 units of the H200 from Nvidia.

The H200 is an older Nvidia chip not restricted by current US export controls, unlike newer processors that remain banned on national security grounds.

The company’s stock had previously dropped over 20% from its October peak due to concerns about ecommerce earnings and regulatory scrutiny of livestreaming in China.

The potential approval is expected to help Chinese firms access AI semiconductors, which are needed to upgrade and run advanced models.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

The H200 may not be the answer China’s AI labs hope for

The Trump administration cleared licensed H200 exports for “approved customers” in China 1. Enforcement is unpredictable, as a smuggling network moving H100 and H200 GPUs was shut down, with over $50 million seized 2.
Built In, citing Lemurian Labs CEO Jay Dawani (Lemurian Labs is an AI chip startup), puts 75% of AI training chips on Nvidia’s CUDA platform (CUDA is Nvidia’s proprietary software platform and programming model for GPUs) 3. Switching to domestic options would require extensive software reengineering.
The H200 features 141 GB of High Bandwidth Memory (HBM) 3e and up to 4.8 TB/s memory bandwidth 4. Any China-bound H200 exports still require licensing and end‑user vetting 1. U.S. officials put China’s AI capability 3–6 months behind the U.S. (per venture capitalist David Sacks) 3.

Data center operators near China should prepare for GPU-grade cooling demand

An Alibaba order of 200,000+ H200s would spike demand in Hong Kong data centers, where providers are adding liquid cooling and 30MW of capacity sits available 5.
Liquid cooling adoption in Hong Kong is early 6. Vendors and integrators deploy direct-to-chip systems (circulating coolant to cold plates mounted directly on processors) that handle ~100°C heat from GPU clusters 7.
Colocation providers (third-party data centers that rent power, space, and cooling) can serve Chinese enterprises needing cross‑border AI compute 5. Global Switch’s Tseung Kwan O facility sits near submarine cable landing stations and the city’s financial and business districts 5.