The Zhitong Finance App learned that Hong Kong AI computing power stocks rose strongly. As of press release, MINIMAX-W (00100) rose 7.32% to HK$352; Smart Spectrum (02513) rose 6.04% to HK$1,244; Guangdong-Hong Kong Bay Smart Computing (01396) rose 6.03% to HK$10.9; and World Data - SW (09698) rose 5.06% to HK$32.
According to the news, DeepSeek plans to increase the overall pricing of the DeepSeek API service in the near future. GF Securities believes that unlike previous price adjustments for peak periods, this time points to the overall pricing of the API service, mainly due to increased demand and tight supply of computing power due to the growth of AI applications. This price increase reflects an adjustment in the price war situation in the big model industry, and manufacturers are paying more and more attention to cost control and sustainable profitability. On the computing power side, the rise of large domestic models is directly driving domestic computing power demand. Domestic AI chips and computing power leasing are beneficial directions.
It is worth mentioning that according to the recent Guangdong-Hong Kong Bay Intelligent Computing announcement, its subsidiary Shenzhen Hongrui and SPDB Financial Leasing signed two financial lease agreements, with a total financing amount of about 1,185 billion yuan, of which the latest amount was about 790 million yuan, and the lease period was 62 months. Guangdong-Gangwan Intelligent Computing signed a new contract of 15 billion yuan in the first half of the year (total order amount exceeding 30 billion yuan) and has already delivered more than 2 billion yuan of equipment, and has completed procurement and delivery of more than 2 billion yuan of equipment. There are plenty of orders in hand, and the pace of delivery continues to accelerate. At a time when the global supply of AI computing power is tight and demand is bursting out, whoever can turn orders into operable computing power assets as quickly as possible can seize the opportunity.
According to CITIC Construction Investment's opinion, the current model competition is shifting from single-parameter expansion to collaborative evolution of pre-training, enhanced learning, inference time calculation, RSI and long-term agent capabilities. Computing power requirements will expand from training to inference and agent execution, and continue to be optimistic about Dachang Capex ecology, domestic chips and supernodes, computing power services, pre-AI, B-side AI applications, and local reasoning.