A New Milestone for Domestic GPUs: The New Landscape of China-US AI Computing Power Competition from Moore Thread's IPO

December 5, 2025, the Shanghai Stock Exchange's STAR Market welcomed the "first domestic GPU stock," Moore Thread Intelligent Technology, with stock code 688795. Its IPO price was 114.28 yuan per share. On the first day, its stock price soared 468%, pushing its market capitalization past 300 billion yuan.

A few days later, relevant data showed the stock's closing price had surged 723.49% compared to the IPO price. Behind this staggering figure lies not only the enthusiasm of the capital markets but also the urgent demand for autonomous control in China's AI computing power sector.

From Moore Thread's IPO to the advancement of China's intelligent computing GPU R&D capabilities, and further to the strategic competition between China and the US in the AI field, a clear line of technological competition is forming, with GPU computing power being the core fuel of this race.

01 A New Starting Point for Domestic GPUs

December 5, 2025, marked a symbolic event for China's tech industry – Moore Thread Intelligent Technology officially debuted on the STAR Market, stock code 688795. Founded in 2020 by Zhang Jianzhong, former global vice president of NVIDIA, the company's backers include prominent institutions such as Sequoia Capital China, Tencent, and ByteDance.

On its first trading day, the stock price skyrocketed 468%, driving its market capitalization to 305.5 billion yuan, making it the largest IPO on the STAR Market since 2024. By December 11th, the stock price had risen 723.49% from the IPO price. This market reaction far exceeded expectations, reflecting investors' high anticipation for the prospects of China's independent GPU development and strong confidence in domestic AI computing power solutions.

Moore Thread's development path and product portfolio reflect the typical strategy of Chinese GPU companies. The company adopts an all-functional GPU strategy, with business covering multiple fields including AI intelligent computing, professional graphics acceleration, and desktop-grade graphics cards. In AI intelligent computing, Moore Thread has released several computing cards like the MTT S4000, whose provided GPU computing power can already meet the demands of large-scale model training. It also launched the "Kuae (KUAE)" full-stack intelligent computing center solution, which integrates computing power leasing services.

It is worth noting that Moore Thread's MTT S80 gaming graphics card became the only domestic GPU capable of running "Black Myth: Wukong" smoothly. This achievement is significant not only in the gaming sector but also demonstrates the technical capabilities of domestic GPUs in graphics rendering and handling complex scenarios, laying the foundation for diversified GPU computing power applications.

02 The Rising Force of China's Intelligent Computing GPUs

Alongside Moore Thread's IPO, China's GPU industry is showing a pattern of rapid development by the "Four Little Dragons." Besides Moore Thread, Biren Technology, Enflame Technology, and MetaX Integrated Circuits constitute the core force of domestic GPUs, committed to enhancing China's independent AI computing power level.

These companies collectively showcased their latest achievements at the 2025 World Artificial Intelligence Conference, covering areas such as AI computing, intelligent computing centers, and optical interconnect GPUs. Biren Technology, in collaboration with several enterprises, released China's first optical interconnect and optical switching GPU super-node – LightSphere X, aimed at providing higher-efficiency GPU computing power cluster solutions.

Enflame Technology showcased the diversified deployment results of its intelligent computing centers in Qingyang, Wuxi, and Yichang. Its launched computing power leasing services are becoming an important way for small and medium-sized enterprises to access AI computing power. The company highlighted the large-scale commercial application of its "Enflame® S60" AI inference card in the pan-internet field, proving that domestic GPU computing power already possesses the capability to support actual business operations.

MetaX Integrated Circuits publicly displayed its latest developed training-inference integrated GPU, Xiyun C600, for the first time. This chip is specifically optimized for the characteristics of the Chinese market's AI computing power demands. With the technological progress of these companies, the domestic computing power leasing market is gradually reducing its dependence on imported hardware.

From the perspective of financing and IPO progress, Chinese GPU companies are accelerating their capitalization. MetaX successfully passed the listing committee review on October 24, 2025, planning to raise 3.904 billion yuan for high-performance GPU R&D and industrialization. Biren Technology and Enflame Technology have also entered the IPO tutoring stage.

Behind this development trend is the urgent demand of the Chinese market for autonomous computing power. NVIDIA CEO Jensen Huang recently confirmed in an interview that its market share in China had dropped from a peak of 95% to 0%. Although this change is partly due to external environmental factors, it also provides a rare market opportunity for local Chinese GPU companies, especially in the field of computing power leasing services, where the space for domestic substitution is vast.

03 The Competitive Advantages of American GPU Manufacturers

In the global AI computing power chip market, American companies still dominate. Among the world's top 10 AI chip companies in 2025 as selected by research institutions, American firms occupy the vast majority of seats. The GPU computing power solutions they provide remain the backbone of global AI infrastructure.

NVIDIA has consolidated its leading position in AI computing power technology with its Hopper and Blackwell architectures. Its GB200 NVL72 leads with a nearly 78% profit margin, primarily due to its leading computing performance and the CUDA ecosystem barrier. NVIDIA's GPU computing power product line covers a complete ecosystem from data centers to edge computing and offers flexible computing power leasing options through cloud service providers.

AMD, as the second-largest competitor by market capitalization, has actively pursued the AI computing power market in recent years, launching the MI300 series to meet AI training demands and introducing the MI350 series in 2025 to compete directly with NVIDIA's H200. During periods of NVIDIA chip shortages, many organizations switched to AMD's alternatives, promoting the formation of a diversified AI computing power supply chain.

Beyond these two giants, the US has a series of emerging companies focused on AI accelerators. Groq, a startup founded by former Google employees, develops LPU-architecture AI chips focusing on large language model inference. Its innovative architecture provides alternative solutions for specific AI computing power scenarios.

SambaNova Systems, founded in 2017, is dedicated to developing high-performance, high-precision integrated software and hardware systems designed specifically for large-scale generative AI workloads. The company developed the SN40L chip and has raised over $1.1 billion. Its solutions also include a computing power leasing business model.

Furthermore, tech giants are actively deploying self-developed chips. Google launched its seventh-generation TPU "Ironwood," specifically designed for large-scale model training, inference, model serving, and agent workflows, further refining the competitive landscape of the global AI computing power market.

04 The Multi-dimensional Competition in the China-US AI Rivalry

The competition between China and the US in the AI field has transcended mere technological competition, evolving into a comprehensive contest covering technology, markets, supply chains, and standard-setting. The core of this competition is the autonomous control and ecosystem building of AI computing power.

The competition over technological ecosystems is particularly critical. NVIDIA's CUDA ecosystem has become the de facto standard for AI development, establishing a strong moat. Chinese GPU companies are striving to build their own software ecosystems but face enormous pressure to catch up. Moore Thread's "Kuae" full-stack intelligent computing center solution and Enflame Technology's "integrated software and hardware" solutions are attempts to build an autonomous AI computing power ecosystem, with computing power leasing services serving as a key strategy to lower the entry barrier for enterprises.

The reshaping of the market landscape is also accelerating. US export controls on chips to China are pushing Chinese AI labs towards domestic alternatives. Huawei's Ascend 910C can achieve 60% of the inference efficiency of NVIDIA's H100 chip. While this shift may impact performance in the short term, in the long run, it will promote the maturity of China's autonomous GPU computing power technology and spur the prosperity of the local computing power leasing market.

Supply chain security has become a focal point for both sides. Chinese GPU companies are striving to build a complete industrial chain from design to manufacturing, reducing dependence on external technologies. Meanwhile, American companies are seeking markets outside China, with the Middle East becoming a new focal point for deployment. The global AI computing power supply chain is trending towards regional segmentation.

Innovation in computing power service models has become a new dimension of competition. With the popularization of the computing power leasing business model, AI computing power is shifting from a heavy-asset investment to flexible service consumption. Leveraging its vast application scenarios and data resources, China is exploring computing power leasing service models suitable for its local market, which may become an important manifestation of latecomer advantages.

The outcome of this contest will profoundly influence the development direction of the global AI industry. China possesses vast data resources and application scenarios, while the US controls advanced technology and a mature ecosystem. The competition and cooperation between the two will jointly shape the distribution pattern and technological path of global AI computing power.

On Moore Thread's first trading day, the constantly changing numbers next to stock code 688795 on the trading hall's electronic screen not only recorded the changing market value of a company but also symbolized China's determination on the path to AI computing power autonomy.

When NVIDIA's CEO admitted its Chinese market share had dropped from 95% to 0%, China's GPU "Four Little Dragons" were showcasing their latest achievements in optical interconnect GPUs and training-inference integrated chips at industry conferences. From the hard metrics competition of GPU computing power to the business model innovation of computing power leasing, the China-US rivalry in the AI computing power domain has entered a comprehensive, multi-dimensional stage.

The future of global AI computing power is being quietly reshaped amidst this dynamic contest. Whether Chinese companies can build an autonomous AI computing power system while innovating more inclusive computing power leasing service models may become a key variable in this long-term competition.