Recently, the Global Computing Consortium (GCC) officially released Top 10 Global Computing Power Trends (2026). The report brings together industry insights, takes “computing power unlocking intelligent value” as its main theme, and focuses on five core dimensions: industry trends, AI applications, open-source frameworks, computing hardware, and infrastructure. It provides future development references for partners across the computing power industry chain. This article offers a quick overview of the report’s key points.

1. Global computing power is growing rapidly, competition is intensifying, and computing power has become a strategic high ground in technological competition

Computing power has gone beyond traditional infrastructure and become the underlying foundation of the digital economy. The report shows that computing power is a multiplier for economic development. Increasing investment in computing power can effectively drive industry innovation and growth. In the first half of 2025, computing power investment contributed approximately 92% to GDP growth in the United States.

Countries are raising computing power to the level of national strategy. China’s 15th Five-Year Plan identifies computing power as a core element of new quality productive forces and proposes that the penetration rate of intelligent applications should exceed 90% by 2030. The United States released its AI Action Plan, while the European Union’s AI Continent Action Plan promotes the construction of AI factories.

In 2025, the combined AI-related capital expenditure of Chinese and U.S. technology giants exceeded USD 500 billion. Hyperscale cloud service providers and internet companies accounted for 60% of global computing power investment. Over the next five years, AI computing power is expected to grow by a thousand times.

2. Large models are becoming the operating system of the intelligent world, and AI is accelerating into all industries

Large models have gone beyond being tools and are becoming the underlying operating system of the intelligent world. The reasoning capability of large models doubles every seven months, while costs are declining exponentially, stimulating a sharp increase in AI demand across industries.

In the first half of 2025, global AI application downloads exceeded 1.7 billion, and total usage time surpassed 15.6 billion hours. As agents take on increasingly complex tasks, demand for computing power is rising by orders of magnitude.

In 2026, AI competition will shift toward value creation, solving real problems, improving industrial efficiency, and enhancing human well-being. The idea that “good AI is AI that brings real benefits” is pushing the industry back toward rationality and will become a core evaluation standard.

3. Crossing the boundary between the virtual and physical worlds, digital intelligence is evolving toward embodied intelligence, with world models becoming key support

Artificial intelligence is moving from the digital world to the physical world, extending from virtual information processing to real-time interaction with the physical world. Embodied intelligence is characterized by having a physical carrier, execution capability, and active interaction ability. Through the closed loop of “perception-cognition-action,” it enables deep interaction with the environment.

As the “brain” of embodied intelligence, world models integrate multi-source data to build internal representations, accurately simulate physical laws, and support environment simulation, action guidance, and decision optimization.

The implementation of advanced embodied intelligence still requires comprehensive innovation across the AI technology stack, hardware systems, and computing architecture. By 2030, the scale of China’s embodied intelligence market is expected to exceed RMB 1 trillion, becoming an important growth engine for the integration of the digital economy and the real economy.

4. Supernodes are becoming the new foundation of computing power, and intelligent computing centers are entering the supernode era

As AI large models surpass the ten-trillion-parameter level and single-task token consumption by agents reaches hundreds of millions, traditional computing architectures are facing bottlenecks.

Relying on technologies such as ultra-high-bandwidth interconnection and unified memory addressing, supernodes weave hundreds or even thousands of AI accelerators into a logically unified high-density computing body. They feature large-scale networking, high-reliability operation, and multi-scenario adaptability, while releasing hardware potential through open software frameworks.

Huawei CloudMatrix384 supernode supports trillion-parameter training and enterprise-level large-scale inference needs. In the future, supernodes will become mainstream infrastructure for high-end intelligent computing centers and lead the transformation of the computing power foundation.

5. Computing architecture is innovating, moving from CPU-centric design to diverse and equal collaboration

The traditional “CPU-centric” master-slave paradigm is being broken. With the explosive growth of AI and the differentiation of computing workloads, equal collaborative architectures composed of diverse processors such as CPUs, GPUs, NPUs, and DPUs have become the core direction of computing architecture evolution.

General-purpose computing is handled by CPUs, AI training relies on the parallel advantages of GPUs and NPUs, and inference prioritizes low-cost ASICs. In the future, the market structure will change accordingly. AI accelerators’ share of total semiconductor spending in data centers will rise to 60%, while the CPU share will fall to 10%. Industrial value will extend upstream and downstream toward system design, interconnection technologies, heterogeneous software stacks, and other areas.

6. Millisecond-level computing power networks are being implemented, making computing power a tradable production factor

The contradiction between soaring computing power demand and resource mismatch is becoming increasingly prominent. Eastern China has concentrated computing power demand but faces high energy consumption pressure, while western China has abundant green energy but relatively limited demand.

“Strengthening computing through networks” has become the core path. China is accelerating the “Eastern Data, Western Computing” strategy and building a three-level latency circle of “1-5-20 milliseconds”: 1ms city network, 5ms regional network, and 20ms national network.

Computing power resources are packaged in a standardized manner, networks gain “computing-aware” capabilities, and orchestration and scheduling enable millisecond-level matching between tasks and resources. As the computing power standards system gradually improves, computing power is becoming inclusive infrastructure like electricity and the internet. Computing power networks will allow computing power to be used on demand, just like electricity.

7. The integration of supercomputing and intelligent computing is building a new paradigm for scientific computing

Frontier challenges such as climate change, controlled nuclear fusion, and new drug development are driving computing demand toward a composite form of “high-precision scientific simulation + efficient AI processing.”

A single supercomputing or intelligent computing architecture can no longer support diverse demands. Supercomputing (HPC) and intelligent computing are moving from independent development toward deep integration, forming a new “supercomputing-intelligent computing integration” paradigm.

The integrated architecture covers the full chain of hardware, software, and systems. At the hardware level, it enables high-speed interconnection among heterogeneous processors and full-precision computing. At the software level, it builds a unified stack compatible with both HPC and AI frameworks. At the system level, it realizes on-demand allocation of computing power through resource pooling and intelligent scheduling.

In fields such as energy exploration and biomedicine, supercomputing-intelligent computing integration has shortened R&D cycles from years to days, breaking through the efficiency bottlenecks of traditional technologies.

8. Open source and openness are becoming the core of the ecosystem, accelerating innovation in the intelligent computing industry

The intelligent computing industry chain shows complex characteristics: high barriers at the foundation layer, strong innovation at the technology layer, and broad coverage at the application layer. No single vendor can meet the innovation needs of all scenarios. Open source and openness have therefore become the core path for aggregating global forces and breaking down technical barriers.

Leading companies are using open source and openness to gather ecosystems and achieve “technology value-added.” Alibaba’s Qwen series models have exceeded 600 million global downloads. DeepSeek promotes “full-stack open source,” driving ecosystem growth. Huawei has announced the full open-sourcing of CANN, covering the needs of developers across all scenarios. In the future, heterogeneous computing chips will become the norm, and platforms such as OpenCL will promote collaboration.

9. Intelligent computing centers are upgrading, with high density, liquid cooling, and clustering becoming core directions

Computing power demand is growing exponentially. GPT-5 training requires 200,000 to 300,000 H100 GPUs, with the cost of a single training run exceeding hundreds of millions of U.S. dollars. This is driving AIDC to leap from the hundred-megawatt level to the gigawatt level.

AIDC is upgrading toward high density, liquid cooling, and clustering. In terms of high density, the power density of a single AIDC rack is increasing rapidly. NVIDIA NVL72 reaches 132kW per rack, and is expected to exceed 650kW by 2027.

Liquid cooling technology is moving from an optional choice to a standard configuration. Cold-plate and immersion liquid cooling are becoming mainstream, reducing PUE to 1.05-1.15.

Clustering has become the core organizational form for unlocking efficiency. Ultra-large-scale computing clusters at the “million-card” level are becoming a necessary condition for large model training. Meanwhile, zero-carbon data center models powered directly by wind and solar energy are emerging, enabling local production and local consumption of renewable energy, and significantly reducing computing power costs and carbon emissions.

10. Quantum computing is becoming engineered, and the commercialization window is arriving

Quantum computing uses quantum superposition and entanglement to achieve efficient parallel computing, and is expected to provide exponential acceleration for complex problems.

Currently, quantum computing is moving from a scientific vision toward the engineering stage. The research focus is shifting from simply pursuing the number of qubits to advancing both scale and quality. The year 2025 will become a turning point from physical qubits to logical qubits.

Technical routes including superconducting, neutral atom, ion trap, and photonic quantum computing are developing in parallel. The next one to two years will be a key window for quantum computing to move from technological breakthroughs to commercial applications. Quantum computing will give rise to disruptive applications in fields such as new drug development and new material design.

In the message of the report, Jin Hai, Chairman of the Global Computing Consortium, wrote: “The speed of breakthroughs in computing power determines the height of progress in intelligent civilization.” Top 10 Global Computing Power Trends (2026) is the crystallization of industry collaboration. It is hoped that this report will provide reference for partners across the computing industry chain, promote computing power innovation that is more inclusive and sustainable, and encourage more practitioners to join the Global Computing Consortium.

The full report has been released on the official website of the Global Computing Consortium (www.gccorg.com). GCC sincerely invites partners from industry, academia, and research to jointly discuss the future of computing power, work together to build an open and innovative computing ecosystem, and support the sustainable development of the digital and intelligent society.