01 Computing Power as a Measure of National Strength

In 2025, the AI field is on the cusp of a new era.

Recently, the U.S. government unveiled an AI infrastructure investment plan named 'Stargate.' Initially, OpenAI, SoftBank, and Oracle will invest $100 billion in Stargate, with this figure expected to rise to $500 billion over the next four years. This marks not only the largest AI infrastructure project in history but also a significant indicator of the U.S. accelerating its global AI technology competition strategy.

Prior to this, the newly elected U.S. President Trump rescinded the AI regulatory order signed by former President Biden, signaling that AI development in the U.S. may enter a more deregulated phase.

Simultaneously, the U.S. government introduced new regulations governing the export of domestically produced AI chips, aiming to impose stringent global export restrictions on U.S.-made AI GPU chips. Based on the deployed chip computing power of various countries and regions, the U.S. will classify them into three tiers, each with distinct sales restrictions. Notably, the third tier, comprising countries and regions such as China, Russia, and Iran that are subject to U.S. arms embargoes, faces a near-total ban on importing AI GPU chips manufactured in the U.S.

In the AI era, computing power stands as the core competitive force. The U.S. Stargate plan aims to expedite the construction of AI computing power infrastructure. Various signs suggest that the global race for AI computing power, particularly between China and the U.S., may intensify further.

02 China's Counterpart to 'Stargate'

As AI model scales expand and applications are implemented, global demand for computing power has exhibited explosive growth.

IDC predicts that the global AI computing market will grow from $19.5 billion in 2022 to $34.66 billion in 2026. Among this, the AIGC computing market is leading the way, with a projected CAGR (Compound Annual Growth Rate) exceeding 90% from 2022 to 2026. Downstream demand will also propel the increase in AI servers, with the global AI server market anticipated to reach $34.7 billion by 2026, accounting for 21.7% of the entire server market. The Chinese market is expected to reach $12.337 billion.

In this race, China has proactively introduced measures to foster the rapid development of AI computing power.

At the end of 2024, the National Integrated Circuit Industry Investment Fund Phase III Co., Ltd. (Big Fund Phase III) participated in the newly established Huaxin Dingxin Fund and Guotou Jixin Fund, contributing 93.1 billion yuan and 71.1 billion yuan, respectively.

Subsequently, the National Artificial Intelligence Industry Investment Fund was established to support the development of the entire AI industry chain.

On January 23, the Bank of China unveiled the "Action Plan for Supporting the Development of the Artificial Intelligence Industry Chain," announcing that it will provide no less than 1 trillion yuan of special comprehensive financial support to various entities across the entire AI industry chain over the next five years. Of this, the total amount of equity and debt financing will be no less than 300 billion yuan. Additionally, a specialized system guarantee compatible with AI technology innovation will be established to cater to the financial needs of all segments of the industry chain.

These initiatives will profoundly impact the further advancement of AI computing power. Sectors such as AI chips, high-end memory chips, and key materials (photoresists, fourth-generation semiconductor materials) are anticipated to benefit significantly.

03 Representative Enterprises in AI Computing Power

Chinese enterprises have increased investments and actively deployed in the AI computing power domain.

The artificial intelligence industry chain map from the Firestone Creation Industry Data Center reveals that the foundational layer of AI primarily comprises hardware devices like AI chips, sensors, servers, and communication equipment, as well as software systems and data. These components provide computing power support.

As of January 24, China boasts over 210,000 enterprises in the foundational layer of AI, with over 60,000 in the hardware sector.

Recommended Representative Enterprises in the AI Computing Power Industry

Huawei: A globally renowned ICT solution provider with primary businesses encompassing communication equipment, intelligent terminals, cloud computing, etc. In the realm of computing power, Huawei has introduced the Ascend series of AI processors and the Atlas AI computing solution.

Hygon: A leader in domestic high-end processors, leveraging AMD's cross-licensing support to establish a product portfolio driven by both CPUs and DCUs. Its Hygon CPU, based on the x86 architecture, excels in the server and workstation markets. The Hygon DCU adopts the international mainstream GPGPU path and CUDA-like architecture, making it the sole commercial acceleration chip in China capable of handling both scientific computing and AI acceleration computing.

Sugon: A prominent player in China's supercomputing industry, leveraging deep technical expertise and extensive resource integration capabilities to comprehensively deploy AI computing power services. Its integrated computing power service platform connects massive heterogeneous computing resources nationwide, weaving an integrated computing power service network. Sugon's AI servers stand out in the market and are widely used in cutting-edge fields such as scientific computing, smart cities, and financial technology.

Loongson Technology: Committed to independent research and development, steadily progressing along the 'Tock + Tick' path. Its latest fourth-generation chip, the 3A6000, offers over 60% improvement in single-core performance compared to the 3A5000, with significantly reduced costs and clear cost-performance advantages. Loongson Technology is fully dedicated to entering the AI chip field, achieving breakthroughs with a CPU + GPGPU fusion solution to infuse powerful computing power into inference AI chips.

Cambricon: A seasoned player in domestic AI chips, making rapid strides since its inception in 2016 and establishing a comprehensive software and hardware product ecosystem spanning the cloud, edge, and terminal. The Thinker 370 smart chip and accelerator card launched in 2021 doubled computing power through Chiplet technology.

InnoSilicon: With a military background and deep roots in the field of graphics display and control, InnoSilicon has robustly entered the domestic GPU civilian market with its independently developed GPU chips. In 2024, InnoSilicon successfully developed the Jinghong series of high-performance intelligent computing modules and complete machine products for AI training, inference, and scientific computing, supporting multi-precision mixed operations and advanced multi-card interconnection technology.

Inspur: A globally leading server vendor with primary businesses including servers, storage, cloud computing, etc. Dedicated to providing professional and efficient computing power platforms.

Computing power remains a long-term theme. Data indicates that as of September 2024, China's total computing power scale reached 246 EFLOPS, with over 13,000 computing power application projects and a total scale of over 8.3 million standard racks of in-use computing power centers, ranking among the world's leaders in terms of total computing power. With the advancement of China's integrated computing power network and the 'East Data, West Compute' project, coupled with successive policy benefits, the domestic computing power industry holds vast prospects. The industry chain will continue to maintain high enthusiasm, and related companies will fully benefit.

—END—

Author | Firestone Creation | Weng Jianping

Review | Firestone Creation | Yin Li