Chinese AI applications go global: Computing power lays the foundation, and scenarios gather strength —— The "Insight Report on the Development Needs of Chinese AI Application Overseas Enterprises in 2025" is released!
In recent years, the global AI market has witnessed explosive growth. According to Bain data, the global market size of AI software and hardware reached $185 billion in 2023 and is expanding at an annual growth rate of 40% - 55%. It is expected to exceed $780 - $990 billion by 2027, with the AI application market size exceeding $407 billion. Against this backdrop, Chinese AI application companies, leveraging technological breakthroughs, domestic scenario innovation experience, and policy support, are accelerating their expansion into overseas markets and becoming important players in the global AI ecosystem.
However, the journey of Chinese AI application companies going global is not without challenges. A survey by 36Kr Research shows that companies face three core challenges: 52.7% of companies have insufficient global computing power infrastructure layout, which directly leads to high latency in cross - regional services and low data collaboration efficiency; 52.0% of companies are troubled by high costs and long cycles of cross - border payment and settlement, which restricts capital turnover and global profitability; 44.3% of companies have a single global marketing channel, making it difficult to break through regional and cultural barriers to achieve accurate customer acquisition. Among them, computing power, as the infrastructure for AI applications, directly affects model training efficiency, inference response speed, and service coverage, and has become a key variable determining the success of going global.
Overall, for Chinese AI application companies to succeed in their overseas expansion, they need to build a comprehensive triangular system of "computing power foundation + marketing breakthrough + payment closed - loop", and the optimization of computing power infrastructure is the primary breakthrough point. Meanwhile, customizing scenario - based infrastructure to meet the personalized needs of different industries and regions and achieve differentiated competition is also the key for companies to gain a foothold in overseas markets.
Against this background, 36Kr Research released the "Insight Report on the Development Needs of Chinese AI Application Companies Going Global in 2025" (hereinafter referred to as the "Report") based on a questionnaire survey of 700 AI application companies going global. The report comprehensively analyzes the current situation, core needs, and future trends of AI application companies going global, focuses on seven popular fields such as AI productivity tools, emotional companionship, audio - video generation, education, games, AI terminals, and embodied intelligence, and dissects the differentiated computing power requirements in the training and inference stages, providing important decision - making basis for companies to formulate overseas expansion strategies.
1. Building a Computing Power Foundation: Globalization, Elasticity, and Scenario - Based Solutions in Parallel
With the widespread popularity of AI applications globally, companies' demand for computing power has shown explosive growth. The report shows that over 70% of companies' computing power investment accounts for more than 10% of their R & D expenditure, and the annual growth of inference demand exceeds 70%. This data fully demonstrates that computing power has become the "primary necessity" driving the development of AI application companies.
However, while the demand for computing power is surging, AI application companies also face many challenges. High global access latency (58.7%) may cause problems such as freezing and slow response when users use AI applications, seriously affecting the user experience; inefficient cross - regional data collaboration (57.0%) restricts companies' ability to integrate resources and optimize business processes globally; the lack of elastic computing power mobilization ability (52.3%) makes it difficult for companies to flexibly respond to large fluctuations in traffic, and it is easy to have situations of resource idling or overload; the pressure of computing power costs (42.3%) is also a real problem that companies have to face, as high computing power costs directly compress companies' profit margins.
To address these challenges, AI application companies have actively explored computing power solutions. GPU cloud, with its features of rapid deployment, flexible expansion, and pay - as - you - go, has become an important layout direction for AI application companies going global. Currently, 87% of companies rely on GPU cloud to support their overseas business, and only 1.0% do not use GPU cloud services. Companies choose GPU cloud essentially to solve the deployment problem of inference with "cloud - based computing power". Among them, 60.0% of companies value the cluster management and resource scheduling capabilities of the cloud platform to deal with cross - regional loads through automated computing power allocation; 51.0% aim at global node coverage to achieve low - latency deployment in multiple regions, matching the real - time interaction needs of users in the inference stage. It can be seen that GPU cloud has evolved from a simple computing power resource to a comprehensive solution for AI applications going global in terms of "technical collaboration + cost optimization + global adaptation" and has become the core path for AI applications to break through the computing power bottleneck when going global.
When Chinese AI application companies going global choose computing power infrastructure service providers, they comprehensively consider multiple factors such as cost, technical support, efficiency, and compliance. Survey data shows that more than half of the companies pay the most attention to cost competitiveness (59.6%), technical support and operation and maintenance guarantee (58.7%), and product and service delivery efficiency (58.3%), reflecting companies' pursuit of a balance between cost reduction and efficient response in computing power investment; the needs for global computing power layout (45.3%) and compliance certification (31.0%) highlight the adaptability requirements of computing power services in different market environments.
Meanwhile, a very noteworthy data point is that in the selection of GPU cloud providers, besides Alibaba Cloud, Google Cloud, and AWS, new competitors are gradually emerging in the market. GMI Cloud, with a favorability rate of 36.3%, has become the third choice for AI application companies going global. The enterprise samples for this data cover multiple mainstream overseas regional markets around the world, and the enterprise types include start - up AI application companies and the overseas AI application businesses of mature multinational technology companies, which can relatively truly reflect the industry's overall recognition of GPU cloud providers.
Scenario customization is also an important direction for computing power solutions. The report's analysis of seven popular fields shows that although these fields have reached a core technical consensus on "low - latency interaction, high - elasticity scheduling, heterogeneous computing power collaboration, and global compliance", there are differentiated needs due to the characteristics of different scenarios. In the training stage, AI productivity tools focus on multi - modal fusion, emotional companionship emphasizes cross - cultural emotion recognition and persona fine - tuning, audio - video generation requires multi - device compatibility and small - language data enhancement, education focuses on cross - age group model adaptation and small - language annotation, games emphasize high - fidelity rendering and cross - platform debugging, AI terminals focus on multi - device collaboration and model lightweighting, and embodied intelligence focuses on safe and compliant simulation and sensor data annotation; in the inference stage, AI productivity tools pursue low - latency real - time interaction, emotional companionship requires 7×24 - hour dynamic response and low - power optimization, audio - video generation values real - time stream processing and sudden traffic expansion, education focuses on real - time teaching interaction and compliance review, games emphasize millisecond - level real - time interaction and graphics computing power optimization, AI terminals need millisecond - level response at the edge and edge - cloud collaborative inference, and embodied intelligence focuses on microsecond - level motion control and real - time decision - making at the edge. For these specific scenario needs, companies need customized computing power solutions to ensure the efficient operation of AI applications.
In terms of cost optimization, technical means such as mixed - precision inference and heterogeneous computing power collaboration are gradually being applied. Through these technologies, companies can significantly reduce computing power costs while ensuring computing power performance, maximizing economic benefits.
2. Marketing Breakthrough: Driven by Social Media, Empowered by AI, and Deeply Rooted in Localization
In overseas marketing, social media operation (63.0%), partner referral (61.7%), and localized content marketing (60.3%) have become the core channels for companies to acquire users. Social media platforms, with their large user base and strong communication capabilities, provide companies with a broad marketing space; partner referral helps companies quickly enter the market by leveraging partners' resources and influence in the local market; localized content marketing creates content that meets the needs of local users by deeply understanding local culture and consumption habits, enhancing users' brand recognition.
However, currently, companies still have many gaps in marketing capabilities. High costs of social media operation (64.0%) require companies to invest a large amount of human, material, and financial resources in social media account operation, promotion, and user interaction; the lack of accurate user profiles (57.7%) makes it difficult for companies to accurately target their customer groups, greatly reducing marketing effectiveness; the difficulty in quantifying advertising ROI (57.3%) makes it impossible for companies to accurately evaluate the effectiveness of advertising campaigns and optimize marketing strategies.
To bridge these capability gaps, AI technology is gradually becoming a powerful weapon for companies to break through in marketing. 67.7% of companies expect AI to optimize social media sentiment monitoring. Through AI technology, companies can monitor brand - related sentiment dynamics on social media in real - time, quickly detect negative information, and conduct crisis public relations; 57.0% of companies need intelligent advertising placement. AI can conduct accurate advertising placement based on users' behavior data, interests, and preferences, improving advertising conversion rates. In addition, automated multi - language content generation has also become an important means to break through the bottleneck of localized production efficiency. Through AI technology, companies can quickly generate marketing content in multiple languages to meet the needs of users in different regions and improve marketing efficiency.
3. Payment Closed - Loop: Compliance First, Efficiency Upgrade, and Ecosystem Integration
Cross - border payment, as a key link in the commercial closed - loop of companies going global, faces many pain points. Complex compliance reviews (61.3%): Different countries and regions have different payment regulations and policies, and companies need to invest a lot of time and effort to ensure the compliance of payment services; insufficient multi - currency settlement (54.0%) restricts companies from conducting business globally and cannot meet the payment needs of users in different regions; the risk of exchange rate fluctuations (51.7%) also brings uncertainty to companies' capital earnings.
To address these cross - border pain points, companies have clear core needs. 65.0% of companies desire to achieve one - stop compliance management by integrating policies of different countries, simplifying compliance processes, and reducing compliance risks; 57.7% of companies need real - time financial tools, such as exchange rate hedging tools, to stabilize earnings expectations and deal with exchange rate fluctuations; 24.7% of companies focus on local adaptation, supporting mainstream regional payment methods such as e - wallets to improve the user payment experience. Only by meeting these core needs can companies achieve the efficient and stable operation of the payment closed - loop and provide strong support for the smooth progress of business activities.
4. Value Release Guided by Actual Combat Data, Providing Reference Guides for Different Roles
Different from reports that focus on macro - analysis or single - technology paths, the unique value of the "Insight Report on the Development Needs of Chinese AI Application Companies Going Global in 2025" lies in its solid foundation of actual combat. Based on an in - depth survey of 700 front - line AI application companies going global, the report presents not just theoretical desk - top deductions but a systematic review of the real market needs. In addition to detailed data and comprehensive analysis, it also provides differentiated value for different types of readers, including corporate decision - makers, technology R & D personnel, investment institutions, and industry researchers.
For corporate decision - makers, the report reveals the core triangular system of "computing power foundation + marketing breakthrough + payment closed - loop", especially the strategic position of computing power as the primary breakthrough point. Based on the analysis of the actual combat data of 700 companies going global, it provides decision - makers with key trend judgments (such as the penetration rate and core value of GPU cloud, the importance of scenario - based customization) and priorities for resource allocation (such as the proportion of computing power investment, the direction of cost optimization). Meanwhile, the report deeply dissects the differentiated computing power requirements of seven popular AI application fields in the training and inference stages and gives specific suggestions for scenario - based computing power deployment. These front - line insights are crucial decision - making bases for formulating accurate overseas expansion strategies and