Sinolink Securities: AI Computing Power Remains Tight, Domestic Large Models Make Breakthroughs

Stock News12-01

Overseas computing power supply chains showed signs of market cap recovery this week, with the optical communication sector leading gains. Domestic large AI models also achieved breakthroughs. Investors are advised to focus on domestic AI-driven sectors like servers and IDCs, as well as overseas AI-related segments such as servers and optical modules. Sinolink Securities' key insights are as follows:

The overseas computing power supply chain rebounded this week, with optical communications posting significant gains. Google Cloud's AI infrastructure head stated that Google plans to double computing capacity every six months, targeting a 1,000-fold increase over the next 4-5 years, highlighting persistent computing power shortages. This drove strength in Google-related stocks. Meanwhile, OpenAI and AWS continue expanding AI infrastructure—AWS announced an additional 1.3GW computing capacity for U.S. government regions, while Google is negotiating large-scale procurement with Meta and exploring localized "TPU@Premises" solutions. OpenAI revealed plans to convert 220 million paid subscribers by 2030, supporting long-term computing power expansion. The firm remains bullish on the computing power supply chain.

Alibaba confirmed growing computing demand during its earnings call, suggesting potential upward revisions to its three-year 380 billion yuan Capex plan. Its cloud business grew robustly (+34%), with all prior-generation GPUs fully utilized. The Qwen App surpassed 10 million downloads in its first week of public testing, marking the year's strongest AI application growth. Additionally, Singapore's AISG abandoned Meta's model in favor of Alibaba's Qwen architecture, signaling deeper global penetration for China's open-source models.

Domestic large models also made strides: DeepSeekMath-V2 achieved IMO gold medal-level performance using a "generator + verifier" self-evolution framework, bringing open-source models closer to closed-source lab capabilities. ByteDance and ZTE's AI phone, launching in early December with 30,000 units initially, will feature high-authority Agent capabilities, integrating ByteDance's Doubao model and ZTE's hardware expertise. The firm is optimistic about accelerated AI edge-device adoption.

Sector Highlights: - **Servers**: Index rose 4.60% this week but fell 5.45% month-to-date. AWS's 1.3GW AI/HPC data center expansion for government regions represents historic growth, while OpenAI's projected 220 million subscribers may drive sustained training/inference demand. - **Optical Modules**: Index surged 16.46% weekly and 13.56% monthly. Google's reiterated computing power targets imply ongoing TPU cluster expansion, benefiting suppliers like InnoLight. - **IDCs**: Index gained 5.58% weekly but dipped 2.53% monthly. Alibaba's Capex potential and Qwen's traction suggest rising inference-side demand, while DeepSeekMath-V2's breakthrough could spur training needs, accelerating domestic IDC builds.

Key Data Updates: - China's telecom revenue grew 0.9% YoY to 1.467 trillion yuan Jan-Oct 2025, with service volume up 9% at constant prices. - October optical module exports fell 11% MoM and 27.6% YoY; Jan-Oct cumulative exports dropped 17%, likely due to overseas factory shifts by domestic firms. - Q3 2025 Capex for Microsoft/Google/Meta/Amazon rose 53%/83%/133%/55% YoY to $16.7B/$24B/$19.6B/$35.1B, respectively.

Risks: AI development delays, U.S.-China tariff volatility, and material shortages.

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Comments

We need your insight to fill this gap
Leave a comment