I. Policy Leadership: Accelerated Implementation of Global Computing Power Strategic Layouts
In 2025, computing power, as the core productive force of the digital economy, has risen to the level of national strategy in various countries. Major global economies have introduced specialized policies to promote the construction of computing power infrastructure and the improvement of the industrial ecosystem.
(I) China: Computing Power Internet Construction Enters a Critical Period of Large-Scale Validation
China's policy layout in the computing power sector continues to deepen, with 2025 becoming a pivotal turning point for the computing power internet transitioning from R&D to industrial implementation.
In May, the Ministry of Industry and Information Technology issued the "Computing Power Interconnection and Interoperability Action Plan", explicitly setting the goal of "achieving an intelligent, real-time-discoverable, and readily-accessible computing power internet by 2028."
In August, the State Council issued the "Opinions on Deepening the Implementation of the 'AI+' Initiative", further emphasizing the need to strengthen intelligent computing power interconnection, interoperability, and supply-demand matching.
Driven by policies, the construction of China's computing power internet test network has made breakthrough progress. The three major telecom operators and the China Academy of Information and Communications Technology jointly launched the computing power internet test network construction, forming a "1 national node, M regional nodes, N industry nodes" ("1+M+N") computing power interconnection and interoperability platform system. By the end of 2025, this system had covered over ten provinces and municipalities including Beijing, Shanghai, and Chongqing, as well as the three major operators' computing power scheduling platforms, cumulatively supporting over 20 million card-hours of service.
(II) United States: Dual Drivers of Chip Act and Cloud Alliance Propel Computing Power Development
The U.S. upgraded its computing power strategy through the National Artificial Intelligence Initiative Act, forming a dual-driver pattern of "Chip Act + Cloud Alliance."
Chip manufacturing reshoring policies drove a 28% increase in domestic advanced process capacity.
AWS, Google, and Microsoft formed the North American Computing Power Alliance to share supercomputing resources.
U.S. investment in quantum computing increased by 42% year-on-year, focusing on software-hardware co-innovation, aiming to take a leading position in the quantum supremacy race.
(III) European Union: Rule-Dominant Layout with Green Computing Power as a Focus
The EU's European Chips Act entered a critical implementation phase, aiming to increase the EU's share of global semiconductor production to 20% by 2030.
Concurrently, the GAIA-X "European Cloud" initiative led by Germany and France completed its technical standards framework, prioritizing the advancement of a green computing power certification system. This system mandates data center PUE values below 1.3, pushing the computing power industry towards a low-carbon direction.
II. Technological Breakthroughs: Computing Power Architecture Innovation and Performance Leap
In 2025, the global computing power technology sector achieved several major breakthroughs spanning chips, servers, quantum computing, and other dimensions, injecting strong momentum into the development of the computing power industry.
(I) Chip Technology: Accelerated Domestic Substitution, Heterogeneous Computing Becomes Mainstream
Quantum Computing Breakthrough: In September 2025, a team from the University of Science and Technology of China successfully developed the photonic quantum computing prototype "Jiuzhang 3.0", achieving precise control over 255 photons for the first time, setting a new world record in quantum computing. When performing Gaussian boson sampling tasks, its speed is one million times faster than "Jiuzhang 2.0" and 120 quadrillion times faster than the world's fastest supercomputer "Frontier".
Accelerated Domestic Substitution for General-Purpose AI Chips: Domestic AI chips have gained capabilities for actual delivery and large-scale deployment in inference computing, specific model adaptation, software-hardware co-design, and system-level optimization. The Xiyun C600, the first fully domestic general-purpose GPU launched by MetaX in 2025, performs between the A100 and H100, realizing a complete domestic supply chain closed-loop from chip design to packaging and testing.
Rise of Inference Chips: As large models entered large-scale commercial deployment, the industry focus shifted from "how to train" to "how to deploy," making inference chips a new market favorite. Google released the inference-specific TPU Ironwood, and NVIDIA reached a deal with inference chip company Groq, signaling that inference is no longer just an adjunct to training. Leading domestic inference chip companies like Intellifusion are building a "new foundation" for China's AI industry with self-controlled underlying technology, ensuring stable, affordable, and sustainable computing.
(II) Server Technology: Ultra-Nodes Become the Focus of Competition
In 2025, the server industry iterated and upgraded around four directions: "high density, energy efficiency, scalability, and intelligence," with ultra-node technology becoming the competitive focus.
Huawei's Ascend 384 Ultra-Node: Broke the CPU-centric von Neumann architecture, innovatively proposing a peer-to-peer computing architecture, achieving major breakthroughs with high-speed bus interconnect technology.
Sugon's scaleX 10,000-Card Ultra-Cluster: Became an industry benchmark as the world's first single-rack 640-card ultra-node. Using a "one-drag-two" high-density architecture design, it achieved ultra-high-speed bus interconnection within a single rack of 640 cards, increasing computing density per rack by 20 times.
ZTE's Orthogonal Ultra-Node System: Built on its self-developed AI high-capacity switching chip, it innovatively launched a system for high-performance training and inference, effectively overcoming the "bottleneck" of traditional single-machine computing power.
(III) Computing Power Interconnection Technology: New Breakthroughs in Cross-Domain Hybrid Training
The Shanghai Artificial Intelligence Innovation Center's "DeepLink Ultra-Large-Scale Cross-Domain Hybrid Training Technology Solution" tackled challenges such as large-scale cross-domain heterogeneous cluster scheduling, integration of high-performance communication protocols, and design of high-reliability fault-tolerance mechanisms.
It achieved a 1,500-kilometer interconnection between AI computing centers in Shanghai and Jinan, completing hybrid training of hundred-billion-parameter large models with equivalent computing power exceeding 95% of that of a single-chip, single-cluster setup.
It also realized interconnection and large-model hybrid training among AI computing centers in Beijing, Shanghai, Guizhou, and other locations, exploring a new path for nationwide intelligent computing power interconnection and the efficient utilization of dispersed computing resources.
III. Market Explosion: Computing Power Leasing Becomes a New Industry Trend
In 2025, the global computing power market experienced explosive growth. Computing power leasing, as an elastic and specialized computing power solution, became a market favorite, with the industry scale expanding rapidly.
(I) Rapid Market Growth
Global Market: The global computing power leasing market size is projected to exceed $180 billion in 2025, representing a year-on-year increase of over 35% compared to 2024.
Chinese Market: The market size is expected to surpass 2,116 billion RMB in 2025, maintaining a high growth rate of over 20%. Intelligent computing power leasing is the core growth driver, with the Chinese intelligent computing power leasing market reaching 377 EFlops in 2025, a 88.5% year-on-year increase.
(II) Diversified Demand Structure
Computing power leasing demand shows a multipolar trend.
Regional Structure: North America leads, holding 42% market share; the Asia-Pacific region is the growth engine with a CAGR of 48%; demand for green computing power is rising significantly in Europe.
Client Structure: Large tech companies contribute 55% of revenue; adoption rate among SMEs jumped from 28% in 2024 to 41% in 2025.
(III) Expanding Application Scenarios
Artificial Intelligence (65%): Covers training and inference for NLP, computer vision, and multimodal models.
Scientific Research (18%): Involves HPC for climate simulation, drug discovery, and astrophysics.
Metaverse & Digital Twins (12%): Industrial design and smart city construction are key implementation areas.
Edge Computing & IoT: Spurred new distributed computing power leasing models. The number of edge computing nodes is expected to reach 350,000 in 2025.
(IV) Gradually Clarifying Industry Landscape
The global computing power leasing industry has formed a three-tier competitive landscape:
Tier 1 (~70%): Cloud service providers and major telecom operators, offering standardized services leveraging their own data centers and network bandwidth.
Tier 2 (~25%): Third-party IDC service providers and equipment vendors, focusing on vertical sectors with resource accessibility advantages.
Tier 3 (~5%): Cross-sector entrants, rapidly deploying via asset-light, customized services.
The competitive focus is shifting from resource grabbing to building technical barriers, penetrating verticals, upgrading operational models, and cost control, moving the industry from "land grabbing" to refined operations.
IV. Future Outlook: The Computing Power Industry Enters a New Development Stage
(I) Computing Power Interconnection Forms Networks, Enabling "Nationwide Integrated Scheduling"
In the future, the computing power internet will continue to deepen its enabling role in industries like education, manufacturing, and finance, further breaking down geographical and sectoral barriers, driving computing power resources towards higher efficiency, broader accessibility, and greener development.
(II) Technology Innovation Drives Continuous Performance Improvement
Chip Technology: Will continue towards specialization and heterogeneity. Chiplet technology will gradually mature, providing core support for post-Moore's Law computing power growth.
Quantum Computing: Will transition from scientific vision to engineering. The research focus will shift from merely pursuing qubit count to "advancing both scale and quality."
Liquid Cooling: Adoption rates will increase, energy efficiency will continue to optimize, and scheduling algorithms will upgrade, potentially pushing cluster utilization beyond higher technical thresholds.
(III) Computing Power Leasing Market Continues to Expand, with Proliferating Model Innovations
The market size will keep growing, with demand from vertical industries growing at over 50% and consumption by SMEs becoming a major growth driver.
Business models will innovate continuously. Pay-as-you-go will dominate, complemented by long-term contracts and reserved instances for stable demand. Mechanisms like computing power futures trading and computing power mortgage financing will mature, and hybrid cloud scheduling platforms will enable seamless integration of private and leased resources.
(IV) Green Computing Power Becomes a Mainstream Trend
As global attention to climate change intensifies, green computing power will become a mainstream trend.
Data centers will accelerate the adoption of the "liquid cooling + green electricity" model to reduce energy consumption and increase renewable energy use.
Countries will strengthen the construction of green computing power certification systems, steering the industry towards low-carbon, sustainable development.
V. Conclusion
2025 was a pivotal year for the global computing power industry. Policy leadership, technological breakthroughs, and market explosion jointly propelled the industry into a new stage of development. As a vital component, computing power leasing plays a crucial role in meeting enterprises' elastic computing demands and promoting the accessibility of computing power. Looking ahead, with the continuous maturation of interconnection technologies, the ongoing emergence of innovations, and the further release of market demand, the global computing power industry will embrace even broader development prospects, providing solid support for the thriving development of the digital economy.
--This report is compiled by OmniYQ Cloud Computing Platform.