In recent years, the global artificial intelligence market has expanded rapidly. According to Bain & Company, the global AI market, including related hardware and services, reached USD 185 billion in 2023, and is growing at an annual rate of 40–55%. By 2027, the market is expected to surpass USD 780–990 billion, with AI applications alone accounting for over USD 407 billion. Amid this growth, Chinese AI application companies are accelerating their international push, leveraging advancements across AI generations, domestic experience with novel use cases, and strong policy support. They are emerging as influential players in the global AI ecosystem.
Yet, this expansion is not without obstacles. According to 36Kr Research Institute, three structural barriers exist:
- 52.7% of companies report a lack of global computing infrastructure deployment, leading to latency issues and inefficient cross-border data collaboration.
- 52.0% cite high costs and long processing times for international payments, constraining cash flow.
- 44.3% struggle with limited global marketing channels, hindering cultural adaptation and user acquisition.
Among these, computing infrastructure—the backbone of AI deployment—stands out as the most immediate challenge, affecting model training efficiency, inference response speed, and service coverage.
To overcome these hurdles, companies need to establish a closed loop of computing infrastructure, market penetration, and monetization. Among these, computing infrastructure represents the primary breakthrough point. At the same time, delivering customized, scenario-driven solutions tailored to local needs will be essential to achieve a solid foothold in overseas markets.
In response to these challenges, 36Kr Research Institute surveyed 700 Chinese AI application companies expanding globally, and released the report titled “AI Applications Expanding Globally: 2025 Insight Report on Infrastructure, Market Readiness, and Scenario-based Differentiation.”
The report analyzes the current state, core demands, and emerging trends shaping the global expansion of those companies. It focuses on seven high-growth segments: AI productivity tools, emotional companions, audio and video generation, education, gaming, AI terminals, and embodied intelligence, unpacking their varied infrastructure needs across both training and inference stages. These insights aim to provide a practical foundation for companies developing global expansion strategies.
1. Infrastructure: Elastic, global, and scenario-driven
As AI applications become more widely adopted, demand for computing infrastructure is surging. According to the report, over 70% of surveyed companies allocate more than 10% of their R&D budgets to computing resources. Inference-related demand alone is growing by more than 70% annually. These figures underscore that computing infrastructure is now among the most urgent growth drivers for AI firms.
However, this demand has also brought a host of challenges.
- High latency (58.7%) degrades user experience through lag and slow response times.
- Low efficiency in data collaboration (57%) hinders ability to integrate global resources and streamline operations across regions.
- Insufficient capacity for computing resource orchestration (52.3%) makes it difficult to manage sudden spikes in user traffic.
- Cost pressures related to computing infrastructure (42.3%) squeeze already thin profit margins.
To address these challenges, companies are actively exploring new solutions, including GPU cloud services as they are efficient to deploy, easy to scale, and flexible in pricing. Currently, 87% of surveyed firms use GPU cloud platforms to support their international operations, while only 1% have yet to adopt them.
GPU cloud platforms are gaining traction primarily for their ability to support inference with scalable cloud-based computing. Key factors include cluster management (60%), global node coverage (51%), and dynamic traffic handling. This highlights the expansion of GPU cloud solutions from a form of computing resource into an amalgamation of technical coordination, cost optimization, and global adaptability—critical to overcoming AI-related computing bottlenecks.
In choosing infrastructure providers, Chinese AI companies consider a combination of factors. Survey data shows that 59.6% of companies prioritize cost competitiveness, 58.7% value technical support and O&M (operations and maintenance) guarantee, and 58.3% focus on product and service delivery efficiency. Demand for global resource deployment (45.3%) and compliance certifications (31%) also reflect the need for regional adaptability.
Notably new providers are beginning to gain traction alongside major players like Alibaba Cloud, Google Cloud, and Amazon Web Services (AWS). GMI Cloud was the third most preferred provider, with 36.3% of surveyed companies including it in their evaluation. GMI Cloud’s rise suggests ongoing shifts in provider dynamics across core international markets.
Scenario-based customization has emerged as a critical differentiator, though infrastructure needs differ widely across use cases.
During the training phase:
- AI productivity tools focus on multimodal fusion.
- Emotional companions emphasize cross-cultural emotion recognition and persona fine-tuning.
- Audio and video generation models require multi-device compatibility and low-resource language data augmentation.
- Education applications prioritize cross-age model adaptation and annotation in underrepresented languages.
- Gaming demands high-fidelity rendering and cross-platform debugging.
- AI terminals stress multi-device collaboration and model lightweighting.
- Embodied AI targets safe, compliant simulation environments and sensor data annotation.
During the inference phase:
- AI productivity tools require low-latency, real-time interaction.
- Emotional companions need round-the-clock dynamic responses with low-power optimization.
- Audio and video generation models prioritize real-time stream processing and burst traffic scalability.
- Education applications shift focus to real-time teaching interaction and compliance auditing.
- Gaming requires millisecond-level interaction and optimization of graphical performance.
- AI terminals need support for millisecond-level on-device response and device-cloud collaborative inference.
- Embodied AI emphasizes microsecond-level motion control and on-device decision-making.
To meet these diverse requirements, companies should develop tailored infrastructure strategies.Technologies such as mixed-precision inference and heterogeneous computing resources collaboration are being increasingly adopted to reduce costs, maximizing economic efficiency without compromising on performance.
2. Market strategy: AI-enabled, platform-driven, and locally tailored
To acquire users abroad, Chinese AI firms primarily rely on social media operations (63%), channel partnerships (61.7%), and localized content marketing (60.3%). These channels offer scale, trust, and cultural resonance, but can come with steep costs and technical barriers.
However, companies may still encounter capability gaps in global marketing:
- Poor audience segmentation (57.7%) hinders the ability to identify and target key customer segments, reducing the effectiveness of marketing efforts.
- Meanwhile, limited visibility into advertising ROI (57.3%) prevents accurate assessment of ad placement impact, making it difficult to optimize strategies accordingly.
- Social media operations remain costly for 64% of firms, as they require substantial investment in manpower, tools, and capital.
AI technologies can help fill these gaps. 67.7% of companies anticipate leveraging AI to monitor social media sentiment in real time, enabling timely detection of marketing signals and potential crises. 57% said they would apply AI to optimize ad targeting based on user behavior.
In addition, automated multilingual content generation has become critical to overcoming bottlenecks in localized content production. Using AI, companies expect more efficient creation of marketing content that better meet the needs of diverse audiences.
3. Monetization: Compliance-first, localized, and risk-aware
While critical to international expansion, cross-border payments remain fraught with challenges. Complex compliance reviews (61.3%) consume substantial time and resources. Limited multi-currency settlement capabilities (54.0%) could restrict companies’ ability to operate globally due to incompatibility with payment preferences of users across different regions. Exchange rate volatility (51.7%) also introduces uncertainty to revenue streams.
In response to these issues, companies have outlined several core requirements.
- 65.0% seek one-stop compliance management to streamline regulatory processes and reduce risk by integrating policies across jurisdictions.
- 57.7% require real-time financial instruments, such as foreign exchange hedging, to stabilize revenue expectations and manage currency volatility.
- 24.7% prioritize localized payment options, including support for regional mainstream payment methods like digital wallets, to enhance user payment experiences.
Meeting these needs is critical to establishing a reliable, efficient cross-border payment loop that supports long-term global operations.
4. A practical blueprint, grounded in frontline data for diverse stakeholders
Unlike reports limited to macro-level analysis or narrow technical trajectories
Going beyond macro-level analysis and technical summaries, the “AI Applications Expanding Globally: 2025 Insight Report on Infrastructure, Market Readiness, and Scenario-based Differentiation” delivers field-driven insights grounded in real-world conditions. Based on in-depth research involving 700 Chinese AI application firms expanding globally, it offers a systematic overview of actual market demands and actionable strategies.
For corporate decision-makers, the report outlines a closed loop of computing infrastructure, market penetration, and monetization, identifying infrastructure as the primary point of leverage. Drawing on real-world data, it highlights trends such as GPU cloud adoption, the rising importance of scenario-specific customization, and the priorities shaping computing investments and cost strategies. It also provides an in-depth breakdown of differentiated infrastructure demands across the training and inference phases in seven high-growth segments, along with practical recommendations for resource deployment. These insights may serve as a foundation for crafting targeted global strategies and managing key risks.
For AI R&D and engineering teams, the report offers highly adaptable, technically specific guidance. It maps sector-by-sector challenges to recommended infrastructure configurations and optimization strategies. It also explores trends such as scenario-defined infrastructure, software-driven hardware value realization, and multimodal fusion, offering practical input for tech selection and system architecture enhancement.
For investors and industry researchers, the report delivers a panoramic view of China’s AI global expansion. It quantifies market opportunities, pinpoints common challenges, and reveals the paths taken by leading players. The in-depth analysis of divergent development trajectories and computing demands across seven key segments offers valuable benchmarks for identifying high-growth opportunities, evaluating technical barriers, and assessing business model sustainability.
The report’s exploration of infrastructure optimization strategies, including solutions from emerging providers such as GMI Cloud, also highlights the service layers that underpin the complexities of global expansion.
Notably, AI is evolving at a much faster pace than traditional sectors, and application formats and user expectations can both shift rapidly. The report centers its focus on supporting decision-making over the next one to two years, which represents a crucial window for cross-border expansion. Amid this volatility, demand for efficient, stable, and cost-effective infrastructure will foreseeably remain constant.
Through its detailed assessment of infrastructure-related challenges, evaluation of GPU cloud and other solutions, and emphasis on scenario-specific deployment, the report addresses this core imperative.
Together, these insights offer a durable foundation for global expansion, equipping companies with the clarity to stay competitive even as technologies continue to evolve. Backed by real-world data, they also provide companies the clarity needed to navigate global markets with strategies built to last.
The full report, including raw data and detailed insights into computing requirements across the seven featured sectors, is available for download here.
This article was published in partnership with 36Kr Research Institute.