Summary of Key Points from the Conference Call Industry Overview - The conference call focuses on the **China Internet** industry, particularly the **AI applications** sector and its dynamics in July 2025, highlighting trends in **cloud service providers (CSP)** and **AI model performance**. Core Insights and Arguments 1. **Chip Supply Dynamics**: - The evolving dynamics of Nvidia's H20 chip supply are crucial, with potential resumption of chip sales to China being discussed. This could lead to a significant increase in CSP capital expenditures (capex), projected to rise by **42% quarter-over-quarter in 3Q25** from a likely low in 2Q25 [1][1][1]. 2. **AI Model Launches**: - Continued launches of foundation models are noted, with performance gaps between US and Chinese models narrowing. OpenAI's GPT-5 launch is mentioned, but new models from Chinese platforms like Zhipu's GLM-4.5 and Alibaba's Qwen are showing competitive performance [1][1][1]. 3. **Annual Recurring Revenue (ARR) Trends**: - Monthly ARR trends for popular AI video generation models are highlighted, with **80% of China's AI ARR generated from overseas**, despite only capturing **5% of the total global AI applications revenue**. Key applications include video generation and image editing [1][1][1]. 4. **Engagement Trends**: - There is a noted **6% month-over-month decline** in engagement for consumer-facing AI chatbots in July, attributed to increased integration of AI functions into super-apps. Specific apps like DeepSeek and Doubao saw declines of **10% and 13% month-over-month**, respectively [1][1][1]. 5. **Enterprise AI Adoption**: - The adoption of AI by Chinese enterprises is accelerating, with token usage increasing by **404% and 284% year-over-year** for AI-native apps and in-app AIs, respectively. Notably, **66% of the top 30 AI apps** are developed by major internet companies: Alibaba, Baidu, ByteDance, and Tencent [6][6][6]. 6. **Mobile App Engagement**: - Overall engagement across the top 400 mobile apps increased by **6% year-over-year** in July 2025, with significant growth in Weixin and Douyin app engagement, which grew by **6% and 19% year-over-year**, respectively [7][7][7]. 7. **E-commerce and Local Services**: - E-commerce engagement grew by **14% year-over-year**, with JD and Taobao showing strong growth rates of **76% and 11% year-over-year**. Local services engagement also accelerated to **18% year-over-year** [11][11][11]. 8. **Gaming Engagement**: - Gaming engagement increased by **3% year-over-year** in July, with specific titles like Tencent's DnF mobile maintaining stable time spent shares [10][10][10]. Additional Important Insights - The report emphasizes a more defensive investment strategy due to weaker profit setups in transaction platforms, particularly in e-commerce and local services [10][10][10]. - The competitive landscape for AI applications is evolving, with significant implications for gaming and video generation due to advancements in multi-modal AI models [1][1][1]. - The report includes detailed statistics on the performance of various AI applications, highlighting the competitive positioning of companies like Kuaishou and ByteDance in the AI video generation space [36][36][36]. This summary encapsulates the key points discussed in the conference call, providing insights into the current state and future outlook of the China Internet and AI applications industry.

Nvidia CEO Jensen Huang and U.S. Secretary of the Interior Doug Burgum attend the "Winning the AI Race" Summit in Washington D.C., U.S., July 23, 2025. China has told companies to refrain from using Nvidia's H20 chips after the chipmaker recently received approval to resume shipping the less advanced artificial intelligence product, Bloomberg reported, citing sources familiar with the matter. Authorities have recently told companies to avoid using the Nvidia chips, or those from Advanced Micro Devices, for ...

Core Viewpoint - Nvidia has responded to allegations from Chinese state media regarding its H20 artificial intelligence chips being a national security risk for China, asserting that the claims are unfounded and defending the technology of its chips [1][2]. Group 1: Allegations and Responses - Chinese state media, specifically an account affiliated with CCTV, criticized Nvidia's H20 chips as being neither technologically advanced nor environmentally friendly, suggesting that consumers have the option to avoid purchasing them [1][2]. - The article from Yuyuan Tantian claimed that the H20 chips could have a hardware "backdoor" allowing for remote shutdown capabilities, which Nvidia has firmly denied, stating that cybersecurity is a priority and that there are no such backdoors in their chips [2][3]. - Nvidia has also rejected accusations that its AI chips contain a "kill switch" that could remotely deactivate them, emphasizing the integrity of their technology [2][3]. Group 2: Market Context and Export Controls - The resumption of Nvidia's H20 chip shipments to China marks a reversal of a previous ban imposed by the Trump administration in April, indicating a shift in the company's strategy regarding the Chinese market [3]. - The H20 chips are considered less advanced compared to Nvidia's flagship H100 and B100 chips, and were specifically developed for the Chinese market following initial export restrictions on advanced AI chips in late 2023 [3]. - U.S. export controls on certain Nvidia chips are driven by national security concerns, particularly regarding the potential for Beijing to leverage advanced chips for advantages in AI and military applications [4].

Core Insights - At least $1 billion worth of Nvidia computer chips were smuggled into China following the imposition of export controls by the Trump administration [1] - The B200 chip, favored by major US tech firms for AI applications, is banned for sale to China due to performance threshold regulations [1][5] - Chinese suppliers continued to sell Nvidia chips, including the B200, to data center operators supporting local tech firms despite the export restrictions [2][6] Group 1 - A Chinese data center operator indicated that export controls have not effectively prevented advanced Nvidia products from entering China, instead creating inefficiencies and profits for middlemen [3] - The Trump administration had previously banned Nvidia from selling the less powerful H20 chips, which were designed to comply with earlier export controls [3] - Nvidia's CEO revealed that Trump reversed the ban on H20 chip sales to China, leading to speculation about Chinese companies circumventing export controls [4][7] Group 2 - Evidence reviewed by the Financial Times indicated that Chinese distributors in Guangdong, Zhejiang, and Anhui provinces sold restricted Nvidia chips, including the B200, H100, and H200 [6] - There is no evidence that Nvidia was involved in or aware of the illicit sales to Chinese entities, as the company maintains compliance with US laws [6] - Nvidia stated that assembling data centers from smuggled products is technically and economically unfeasible, emphasizing the need for authorized products and support [8]

Investment Rating - The report maintains a "Buy" rating on PDD, indicating favorable risk-reward dynamics due to its non-participation in the food delivery competition and potential growth from new user traffic and general merchandise strength [10]. Core Insights - The report highlights a potential resumption of Nvidia H20 chip supply, which could positively impact China's cloud service providers' capital expenditures starting from Q3 2025, with an expected 44% quarter-over-quarter increase in aggregate capex [1]. - There is a notable increase in generative AI adoption among Chinese enterprises, with over 40% having pilot-tested generative AI tools, up from 8% last year [1]. - The performance gap between US and Chinese AI models is narrowing, with new releases from various internet platforms and AI startups [1]. - The report emphasizes steady progress in monetization of AI applications, with Chinese models achieving scalable annual recurring revenue (ARR) for their AI products [1]. Summary by Sections AI Application Trends - China's top AI applications showed healthy user engagement trends in June, with a 6% month-over-month increase in domestic AIGC application engagement, driven by strong growth in Doubao and DeepSeek [8][10]. - The overall time spent on the top 400 mobile apps increased by 7% year-over-year in June 2025, with significant growth in eCommerce and social engagement [7][10]. Capital Expenditure Forecasts - The report forecasts a drop in combined capex from China CSPs in Q2 2025, followed by gradual improvement in Q3 and Q4 2025, driven by the resumption of Nvidia shipments and domestic chip ramp-up [15]. Monetization and Revenue Growth - The ARR of various AI operations in Chinese companies shows significant figures, with Kuaishou's Kling AI expected to reach over US$400 million in total annual revenue by 2027 [20][21]. - The report notes that subscription-based productivity tools and advertising-based AI search engines are contributing to the monetization of AI applications [1]. Competitive Landscape - Competition in video-generative models is intensifying, with Kuaishou's Kling AI projected to achieve substantial revenue growth, alongside other models from ByteDance and Alibaba [1][6]. - The report highlights the increasing capabilities of Chinese AI models, which are closing the performance gap with US counterparts [33][34].

Core Insights - Nvidia CEO Jensen Huang expressed optimism about resuming sales of H20 AI chips to China after a three-month pause, highlighting the company's significant market cap of $4 trillion [2][7] - The resumption of chip sales is linked to a broader exchange between the U.S. and China regarding rare earths, indicating geopolitical influences on technology trade [3][7] - Nvidia's market share in China has been nearly halved due to U.S. export controls, resulting in a loss of $2.5 billion in sales during the April quarter and an expected $8 billion hit in the July quarter [9][10] Company Meetings and Discussions - Huang had a productive meeting with Chinese Vice Premier He Lifeng, focusing on topics excluding battery technology and rare earths [4] - Discussions with Xiaomi's CEO Lei Jun revolved around AI applications in large language models, autonomous driving, and robotics, with Xiaomi utilizing Nvidia's automotive chips [5] Export Controls and Market Impact - Nvidia anticipates resuming H20 chip shipments to China following U.S. government assurances, after a halt due to new export requirements [7] - The U.S. government is still processing licenses for Nvidia to sell chips to China, with a potential nine-month timeline to restart the supply chain [11] Competition and Industry Landscape - Huang acknowledged Huawei's capabilities in chip design and AI systems, emphasizing that underestimating Huawei's manufacturing prowess is misguided [12] - Huawei has developed its own AI chips and cloud systems, which do not rely on Nvidia's technology, indicating a competitive landscape for AI development in China [12] Chinese AI Models - Huang praised the quality of Chinese AI models, such as DeepSeek and Qwen, noting their open-source nature and widespread adoption for various applications [14][15] - The emergence of Chinese AI models poses a challenge to established players like OpenAI, particularly in terms of cost efficiency and accessibility [14][15]