Global AI Computing Power Competition Heats Up: Sino-US-European Strategic Layouts and Trillion-Level Market Explosion

Artificial intelligence is reshaping the global landscape at an unprecedented pace, and computing power, as the "new oil" of the AI era, has become the core focus of global technological competition.

The "Stargate" project plans to invest $500 billion to build ultra-large-scale AI computing power clusters; China's daily Token consumption surged from 100 billion in early 2024 to 10 trillion by the end of March 2025, a 100-fold growth in one year; the EU has invested €10 billion to establish 13 artificial intelligence factories.

Behind these numbers lies the unprecedented investment and strategic deployment in AI computing power infrastructure by countries worldwide. From the U.S.'s chip hegemony to China's independent innovation, from Europe's regulatory-first approach to the international cooperation within the Shanghai Cooperation Organization, a global competition around AI computing power is in full swing.

01 Explosive Growth in Computing Power Demand, Token Consumption Soars Exponentially

AI computing power demand is experiencing explosive growth. According to a Huatai Securities research report, Google's Token count doubled from May to July 2025, reaching 960 trillion per month.

By the end of June 2025, China's daily Token volume increased more than 300-fold compared to early 2024, reaching 30 trillion. The significant rise in Token consumption directly reflects the accelerated implementation and scaled deployment of AI applications across various industries.

Global tech giants are actively expanding their computing power infrastructure. Meta is secretly building two "gigawatt-level" supercomputing clusters in Ohio and Louisiana; Musk's xAI Memphis supercluster Phase I is fully operational; Alibaba announced that it will invest over 380 billion yuan in cloud and AI hardware infrastructure construction from 2025 to 2027.

02 Paradigm Shift in Reasoning, Agent Applications Drive New Computing Power Demand

The AI reasoning paradigm is undergoing significant changes. Compared to traditional Chatbots, AI Agents perform task decomposition and coding during execution, leading to a comprehensive increase in interaction frequency, task complexity, and usage frequency.

Huatai Securities analysis suggests that the widespread application of Agents will increase Token consumption tenfold or more, with corresponding computing power demand growing a hundredfold or more. This nonlinear growth relationship stems from the increased complexity brought by multi-Agent collaboration and multi-tool invocation.

In investment research scenarios, research Agents need to acquire data, extract and compute data, read and learn frameworks, and finally generate evaluations, making them far more complex than ordinary dialogue forms. Executing such complex tasks requires stronger computing power support.

03 Global Computing Power Race Intensifies, Varied Strategic Layouts Among Countries

United States: Technological Hegemony and Alliance Building

In January 2025, the Trump administration listed AI competition as a core agenda for national security, overturning the Biden administration's "safety-first" regulatory approach in favor of an "innovation-led" strategy.

The "Winning the AI Race: The U.S. Artificial Intelligence Initiative" revolves around three pillars: accelerating AI innovation, building U.S. AI infrastructure, and leading international AI diplomacy and security. The plan explicitly states the need to "counter Chinese influence in international governance institutions" and "strengthen AI computing export controls."

China: Independent Innovation and Domestic Substitution

Facing U.S. technological restrictions, China is accelerating the process of domestic substitution for AI chips. According to Moore Thread's prospectus data, companies like Cambricon and Hygon have achieved a high level of industrialization in AI chips and have already realized large-scale revenue in the domestic market.

Chinese cloud providers are significantly increasing capital expenditures. Tencent's 2025 capital expenditure is expected to account for a "low double-digit percentage" of annual revenue, potentially approaching hundreds of billions. Huawei Executive Director Wang Tao stated that AI reasoning is experiencing explosive growth, with daily Token counts for generative AI reaching tens of trillions.

European Union: Regulated Development and Sustainable Pathways

In April 2025, the European Commission released the "Continental AI Action Plan," aiming to make the EU a "continent of artificial intelligence." The plan focuses on five key areas: computing power infrastructure, data, applications, talent, and regulation.

The EU plans to establish at least 13 AI factories and up to five AI super factories. It has proposed a cloud computing and AI development bill to promote private investment in cloud computing and data centers, with the goal of at least tripling EU data center capacity in the next five to seven years.

Shanghai Cooperation Organization: International Cooperation and Inclusive Development

In September 2025, the Council of Heads of State of the Shanghai Cooperation Organization (SCO) issued the "Statement on Further Deepening International Cooperation in Artificial Intelligence." The statement emphasizes strengthening cooperation in AI infrastructure, promoting connectivity in AI infrastructure, and advancing collaboration in AI talent cultivation.

Member states will strive to steer AI toward a more open, inclusive, beneficial, fair, and benevolent direction. This is the first comprehensive AI cooperation statement issued by a regional international organization, representing a multipolar trend in global AI governance.

04 Chip Competition Intensifies, Domestic Computing Power Ecosystem Matures Gradually

The global GPU market exhibits an oligopolistic structure. According to Jon Peddie Research data, Nvidia dominates the discrete GPU market, consistently maintaining over 80% market share, while AMD holds nearly 20% of the remaining market.

In 2023, global GPU shipments for intelligent computing centers reached 3.85 million units, a 44.2% increase from 2.67 million units in 2022. Among these, Nvidia's GPU shipments for the intelligent computing center market reached 3.76 million units, accounting for over 90% market share.

China's AI chip market is growing rapidly. According to Moore Thread's prospectus citing Frost & Sullivan data, China's AI chip market size is expected to surge from 142.537 billion yuan in 2024 to 1,336.792 billion yuan by 2029.

The average annual compound growth rate from 2025 to 2029 is 53.7%. GPUs are experiencing the fastest growth rate, with their market share expected to rise from 69.9% in 2024 to 77.3% in 2029.

05 Changes in Computing Power Investment Structure, Intelligent Computing Infrastructure Prioritized

Overall CAPEX among domestic operators has declined, but computing power-related investments are growing against the trend. The three major operators' planned investment scale for 2025 is expected to drop to approximately 289.8 billion yuan, continuing a downward trend.

However, the investment structure is continuously optimizing, with traditional network construction占比 gradually decreasing, while computing power and AI-related investments are achieving逆势增长.

China Mobile expects to invest 37.3 billion yuan in computing power in 2025, accounting for 25% of capital expenditures. The company stated it would dynamically expand reasoning resources based on market demand without setting an investment上限.

China Unicom's 2025 capital expenditure budget is set at approximately 55 billion yuan, but computing power investment is expected to increase by 28% year-on-year, with separate budgets allocated for key AI infrastructure and major projects.

06 Accelerated Application Implementation, Significant Empowerment in Vertical Fields

AI vertical applications are accelerating across various industries. DeepSeek has begun optimizing the research and development process of traditional Chinese medicine, performing computational reasoning on data collected from wearable devices.

The National Children's Medical Center and Beijing Children's Hospital affiliated with Capital Medical University officially released the "Futang Baichuan" pediatric medical large model. With the help of AI models and algorithms, the remote management and control support center of PetroChina Tarim Oilfield has improved alarm accuracy to over 80% and increased management efficiency by more than three times.

Overseas applications are accelerating differentiation, focusing on complex toB scenarios integrated with data. Reviewing the performance of AI applications in the U.S. stock market in 2025, companies with significant stock price gains are those with notable industry Know-how barriers and AI-driven business growth.

Companies like PLTR represent toB enterprises whose strong stock performance core logic lies in AI products deeply empowering their main business, directly driving rapid revenue growth.

07 Future Trends: Multimodal Integration and Ecosystem Competition

Large model technology is accelerating multimodal research and development. In the image field, after GPT-4o's image generation功能 sparked a热潮, Google Gemini2.5FlashImage topped multiple mainstream image benchmarks.

In the video field, Alibaba and ByteDance have refreshed capability heights, while Google has advanced video models to the stage of real-time interactive universal world models.

The competition landscape between closed-source and open-source models is gradually clarifying. The DeepSeek open-source wave promotes model equality and accelerated application, but closed-source models are maintaining performance leadership.

Closed-source vendors like OpenAI and Google still lead with resources, talent, and technological accumulation, while simultaneously pursuing a "open-source + closed-source" collaborative strategy, retaining the advantages of their main models being closed-source while open-sourcing previous generation or lightweight models.

Chip technology iteration is accelerating. Nvidia's chip architecture is transitioning from Blackwell to Rubin; the AMD MI350 series benchmarks against GB200, with the MI400 series planned as the foundation for the new "Helios" server in 2026.

Cloud giants are accelerating the development of self-developed ASIC chips. According to TrendForce, AIServer demand is driving the four major North American CSPs to launch upgraded versions every 1-2 years on average.

The EU plans to at least triple data center capacity in the next five to seven years; the U.S. "Stargate" project plans to invest $500 billion to build ultra-large-scale AI computing power clusters; China's daily Token consumption has grown 100-fold in one year.

The global computing power competition has reached the level of national competition, with AI computing power becoming a basic resource like water and electricity.

Sino-US tech competition, the EU's balance between regulation and innovation, and the SCO's multilateral cooperation—three different development models will shape the global AI landscape.

But regardless of the model, one thing is certain: computing power hegemony will lead the new order of the future intelligent world.

--The above article is purely AI-generated. If there is any infringement, please contact Cloud Engine/Cloud GPU Rental Platform for deletion!