The domestic AI industry is expected to narrow the gap with overseas chains. Western recommends paying attention to companies like Lenovo and Inspur Electronic Information Industry.

Recently, Western released a research report stating that the scissor gap between AI overseas chains and AI domestic chains is still widening. Backtracking to early April, global technology stocks were under pressure due to direct impacts of tariff policies, causing stock prices to fall to temporary lows. However, with the gradual improvement of market sentiment and industry fundamentals, overseas market AI computing power and application-related companies have shown strong rebound momentum. From chips to cloud service providers, to various vertical AI application companies, most companies' stock prices have returned to previous highs, with industry leaders such as NVIDIA and Microsoft even breaking through historical highs. This fully reflects the high recognition of overseas capital markets for AI technology iteration driving industrial change. In contrast, AI industry chain related targets focusing on the domestic market, whether it is basic-level computing chip companies, intermediate-level algorithm service providers, or application-level industry solution companies, have not shown a strong rebound similar to the overseas market. As a result, company stock prices in the AI overseas chain and domestic chain have diverged, forming a scissor gap. Western believes that the performance of the AI domestic chain in the capital market has deviated from the prosperity of the AI industry. Taking DeepSeek as an example, although the R2 model has not been released, the updated version of the DeepSeek-R1 model (R1-0528 version) has shown stronger deep thinking capabilities, performing well in multiple benchmark tests such as mathematics, programming, and general logic, approaching other international top models like o3 and Gemini-2.5-Pro, with notable improvements in complex reasoning tasks. Under Byte's umbrella, the latest 1.6 version of Dou Bao's large model supports deep thinking, multimodal understanding, 256k long- context, graphics-based operations, etc., better supporting the construction of complex agents, while releasing new video generation models and voice podcast models in the multimodal field. Alibaba's large models also show strong iterative momentum, releasing the hybrid reasoning model Qwen-3, integrating fast thinking and slow thinking into the same model. These updates and iterations together confirm the continuous improvement of domestic AI large models in technological cultivation and application landing. By the end of May 2025, the average daily tokens usage exceeded 16.4 trillion, a 137-fold increase from the first release in May last year. In addition to the rapid increase in tokens volume, with the continuous improvement of model capabilities, new applications are constantly emerging. According to the distribution data of tokens in Byte Volcano Engine in May this year and December last year, since the release of deep reasoning models, tokens consumed by AI search have increased tenfold, and AI programming has increased 8.4 times. Additionally, due to the improvement in visual understanding model capabilities, tokens consumption in K12 online education scenarios has increased by 12 times in five months. With the widespread popularity of AI applications and the rapid growth of user numbers, both consumer-oriented applications and enterprise-level solutions require a large amount of cloud computing power to support their efficient operation and real-time response, eventually significantly driving investment in high-performance inference chips and cloud infrastructure, forming a positive "application-computing" cycle. The commercial exploration of domestic AI applications has started slowly, but domestically products with scale income capabilities have emerged in segmented fields. Taking Kuaishou Kling AI as an example, after 10 months of launch (March of this year), the annualized revenue run rate has exceeded $100 million. In April and May of this year, monthly payment amounts exceeded 100 million RMB, with nearly 70% of operating income contributed by P-side (Prosumer, professional users such as self-media video creators and advertising marketers) payment subscription members. This performance also signifies the successful monetization of domestic AI tools in video generation. In addition to the visual content field, in scenarios like emotional companionship (such as PolyBuzz), education (such as Solvely), and intelligent search (such as Quark, Genspark), domestic AI applications also show a strong presence. With the continuous advancement and performance improvement of domestic AI models and the acceleration of AI application monetization processes, the investment value of the AI domestic chain is gradually becoming apparent, and it is recommended to further increase attention and importance to the AI domestic chain. Recommend focusing on: AI computing power domestic chain: AI inference chips (Cambricon, Hygon Information Technology), AI servers (Inspur Electronic Information Industry, Huaqin Technology, LENOVO GROUP), etc. AI applications: TO C: KUAISHOU-W, Wondershare Technology Group, etc., TO B: Hand Enterprise Solutions, Glodon, etc.