Orient Securities released a research report stating that during the 2026 Spring Festival, three major internet giants invested over 40 billion yuan to transition AI from laboratory technology to practical application scenarios, marking a critical point for the widespread adoption of AI applications. Around February, a wave of new model releases emerged both domestically and internationally. Chinese open-source models have become the choice for global developers due to their advantages in performance and pricing. As Anthropic's desktop agent, Cowork, continues to release plugins for various industries, market concerns have grown that AI could bring disruptive impacts to the software industry. While fears of "models consuming software" are reasonable to some extent, the software industry is ultimately expected to undergo a "K-shaped" divergence. The main views of Orient Securities are as follows:

Major players are shifting from model competition to competing over scenarios and ecosystems. During the 2026 Spring Festival, three major internet giants invested over 40 billion yuan to push AI from the lab into real-world applications, reaching a pivotal moment for mass adoption. ByteDance was a key participant, with its Volcano Engine serving as the exclusive AI cloud partner for the CCTV Spring Festival Gala. Its core product, Doubao, featured as the "AI lead" during the gala, recording 1.9 billion AI interactions on New Year's Eve alone. It helped users generate over 50 million Lunar New Year-themed avatars and produced more than 100 million festive greetings. Alibaba's Qianwen launched a "3 billion yuan treat" campaign on February 6, collaborating with ecosystem businesses like Taobao, Hema, and Fliggy to cover users' dining, entertainment, and travel scenarios through free offers. Tencent focused on "social + AI," investing 1 billion yuan starting January 25 to promote the "Yuanbao Pai" feature of its AI assistant "Yuanbao," integrated into WeChat and QQ with an emphasis on social sharing mechanics.

Around February, a wave of new model releases emerged, with domestic models collectively dominating the OpenRouters rankings. On January 27, Moonshot AI released Kimi K2.5, significantly enhancing its Agent capabilities. On February 7 and 14, ByteDance released Seedance 2.0 and Doubao 2.0, respectively; the former gained international attention for its ability to "generate cinema-quality videos in seconds." On February 11, Zhipu AI launched GLM-5, achieving state-of-the-art results in coding and Agent capabilities among open-source models. On February 13, MiniMax released M2.5, which improved the model's ability to break down complex tasks and optimized token consumption during reasoning processes. The advantages of Chinese open-source models in performance and pricing have made them the preferred choice for global developers. According to OpenRouters statistics, over the past week, MiniMax M2.5, Kimi K2.5, and GLM-5 were the three models with the highest token consumption globally, highlighting the comprehensive strength of Chinese models.

Market concerns about "models consuming software" are somewhat justified, but the industry is expected to experience a "K-shaped" divergence. As Anthropic's desktop agent, Cowork, continues to release plugins for various sectors, investors worry that AI could disrupt the software industry, leading to significant adjustments in the U.S. stock software sector. While concerns about "models consuming software" have merit, the software industry will ultimately split in a "K-shaped" pattern: horizontal software lacking unique data resources, deep industry expertise, or network effects will face greater threats from models. In contrast, vertical core systems with exclusive data, low error tolerance, or regulatory lock-in characteristics are at lower risk and are better positioned to integrate AI Agents into their systems to meet customer needs.

Regarding investment targets, against the backdrop of a new round of model iterations and accelerated application promotion by major internet companies, the computing power industry chain presents relatively certain investment opportunities. For AI applications, focus should be on companies with high barriers to entry that are actively transforming. In the AI computing power sector, relevant targets include Haiguang Information, Cambricon Technologies, Runjian Co., Ltd., Runze Technology, UCloud Technology, Capital Online, Hesheng New Materials, Xcener Technology, Dongyangguang, Sugon, and Wangsu Science & Technology. In the AI application sector, relevant targets include Zhuoyi Information, Hehe Information, Transwarp Technology, Shuiyou Co., Ltd., iFlytek, Tonghuashun, and Shiji Information.

Risks include slower-than-expected adoption of AI applications and computing power price increases falling below expectations.