"Domestic GPU Triad" stock prices soar as China's chip industry chain enters an explosive growth period

National Radio and Television Beijing, March 24 (reporter Qi Zhi Ying) shares of Cambricon (688256.SH), known as the “three homegrown GPU leaders,” as well as Moore Threads-U (688795.SH) and Mobaxel-U (688802.SH) have surged collectively. As of the close on March 24, Cambricon’s share price rose 5.89%; Mobaxel rose 13.49%; and Moore Threads rose 8.64%.

Previously, the three companies released performance reports showing that their operating conditions have improved. Experts interviewed said that under the triple drive of “policy + demand + ecosystem,” China’s domestic chips are moving from being “usable” to “really useful,” forming a complete closed loop and a positive feedback cycle spanning chip design, manufacturing, and packaging/testing.

Domestic chips are moving from being “usable” to “really useful”

On March 24, the A-share domestic chip sector saw a strong rebound, closing up 2.83%. Among them, the “three domestic GPU leaders” performed prominently. Cambricon’s share price returned to the 1,000-yuan level, closing at 1028.5 yuan per share, up 5.89%, with an intraday high of 1037.8 yuan per share. Mobaxel’s share price closed at 601 yuan per share, up 13.49%, with an intraday high of 620.44 yuan per share. Moore Threads’ share price closed at 548.96 yuan per share, up 8.64%, with an intraday high of 555 yuan per share.

According to the financial reports released earlier, the year-on-year performance of the three companies in 2025 all improved. In 2025, Cambricon achieved operating revenue of 6.497 billion yuan, up sharply 453.21% year over year; it achieved attributable net profit to shareholders of 2.059 billion yuan, turning from loss to profit, with prior-year attributable net profit of a loss of 452 million yuan. Moore Threads achieved operating revenue of 1.506 billion yuan, up 243.37% year over year; it achieved attributable net profit of -1.024 billion yuan, with year-on-year losses narrowing by 36.7%. In 2025, Mobaxel’s operating revenue was 1.644 billion yuan, up 121.26% year over year; attributable net profit was -781 million yuan, with year-on-year losses narrowing by 44.53%.

Behind the improvement in performance is the explosive growth in the market. With the widespread application of AI and big data technologies, China’s computing power scale has shown a rapid growth trend, and the GPU market scale has experienced explosive growth over the past five years. According to a forecast by Frost & Sullivan, the global GPU market is expected to reach 3.61 trillion yuan by 2029. Among that, China’s GPU market is expected to reach 1.36 trillion yuan by 2029; its expected share in the global market will rise from 15.6% in 2024 to 37.8% in 2029.

“The China chip industry chain is currently undergoing a critical leap from ‘single-point breakthroughs in technology’ to ‘systematic construction of a business closed loop.’ Revenue growth and turning losses into profits—or sharply narrowing losses—represented by the three leading domestic GPU giants are strong evidence of this process.” Tian Lihui, dean of the Financial Development Research Institute at Tianjin University, said that China’s chip industry has moved beyond the early startup stage that relies solely on policy support and has entered a growth stage driven endogenously by market demand and capable of self-sustained “blood replenishment.”

When discussing the pace of domestic substitution in the chip industry, Tian Lihui said it has evolved from “localized acceleration” in the past to “full penetration.” Especially at the strategic commanding height of AI computing power, it is moving from being “usable” to “really useful,” forming a complete closed loop and positive feedback cycle spanning chip design, manufacturing, and packaging/testing.

Jie Jie Micro-Electronics (300623.SH)’s independent director and well-known expert in corporate finance and taxation, Liu Zhigeng, believes that China’s chip industry chain has achieved end-to-end deployment from design, manufacturing, and packaging/testing to materials and equipment, entering a new stage of “linked coordination.” The speed of domestic substitution has moved from accelerating from being “usable” to “really useful.” Especially in 2025, the domestic GPU share in the AI computing power market surpassed 35%, and the substitution process has clearly sped up. There have also been significant improvements in three areas: application scenarios, technological breakthroughs, and an ecosystem closed loop.

A research analyst at West China Securities believes that after the Spring Festival, the logic behind domestic computing power has been further strengthened. With the explosive increase in the number of Tokens and the turning point in the performance of domestic chips, the certainty of the entire computing-power supply chain has increased.

A research analyst at Shanxi Securities said that China’s major customers have continued to increase purchases of domestic AI chips. Meanwhile, domestic first-tier manufacturers are accelerating their catch-up to NVIDIA in areas such as single-card performance, rack-level solutions, ecosystem, and capacity. Their new-generation main products are expected to quickly capture NVIDIA’s chip market share.

A triple drive of “policy + demand + ecosystem”

A research analyst at West China Securities believes that the turning-point in the performance of domestic AI chips validates the impact of the domestic “policy + demand + ecosystem” triple drive on the market. Under the pull from domestic AI demand, leading companies have effectively completed the closed loop for operations, providing certainty and support for the overall industry’s growth potential. At the same time, by promoting the implementation of “artificial intelligence + manufacturing” scenarios through policy, and stacking the drive from computing-power infrastructure, the certainty of orders for domestic computing power has been further enhanced.

In its annual report, Cambricon stated that with the rapid development of artificial intelligence technologies—especially in the fields of large models and general-purpose AI—the demand for computing capability at the underlying-chip level is growing at an unprecedented speed.

An analyst at Oriental Fortune Securities believes that from the demand side, the commercialization models of domestic CSP companies are becoming clearer. AI-related capital expenditures continue to rise. At the same time, domestic models are also being iterated continuously, which is expected to drive an expansion in demand for domestic computing power on the training side.

On the policy side, in January 2026, seven departments including the Ministry of Industry and Information Technology jointly issued the “Implementation Opinions on a Special Action for ‘Artificial Intelligence + Manufacturing’,” clarifying support for breakthroughs in high-end training chips. The Opinions point out to strengthen the supply of AI computing power. Promote coordinated development of intelligent chip hardware and software, and support breakthroughs in key core technologies such as high-end training chips, edge-side inference chips, artificial intelligence servers, high-speed interconnect, and intelligent computing cloud operating systems. Proceed in an orderly manner with the layout of high-level intelligent computing facilities, accelerate the construction of computing-power interconnection and interoperability platforms, the national integrated computing-power network monitoring and dispatch platform, carry out pilot projects for intelligent computing cloud services, promote deployment of large-model all-in-one units, edge-computing servers, and industrial cloud computing power, and enhance the capability to supply intelligent computing resources.

Meanwhile, the MIIT issued the “Action Plan for Integrating the Industrial Internet with Artificial Intelligence for Empowering Industries.” By 2028, the level of industrial internet and artificial intelligence integration for enabling industries will be significantly improved, and at least 50,000 enterprises will be guided to implement upgrades to next-generation industrial networks.

A research report from Shanxi Securities believes that on the supply side, domestic chipmakers represented by Huawei Ascend, Cambricon, Hygon Information, and Kunlun Chip are accelerating their catch-up to NVIDIA in single-card performance. Huawei’s flagship can already be benchmarked against H100, and it is competing with NVIDIA using innovations in super-node architectures and NVIDIA’s rack-level solution approach. At the same time, domestic chips build an ecosystem through two approaches—CUDA compatibility and their own self-developed solutions—gradually breaking through the CUDA ecosystem barrier. In addition, domestic chipmakers are also accelerating solutions to supply-chain issues such as wafer foundry outsourcing, driving faster ramp-up and volume release for domestic chips.

Tian Lihui said that strong domestic demand pulls technology iteration forward. Self-driven technological breakthroughs further improve the bargaining power and security resilience of the supply chain, allowing domestic chip industries to display independent and strong growth momentum during global industry cycle fluctuations.

Experts warn the industry has entered the “elimination race” stage

Liu Zhigeng believes that the main risks for domestic chipmakers at the current stage are concentrated in three areas: overseas technology blockade, reliance on the supply chain, and intensified market competition. First, overseas technology blockades are being continuously strengthened. This not only suppresses development space for areas such as domestic high-performance computing and AI training, but also forces companies to adjust their technology routes under constrained conditions. Second, reliance on the supply chain for key links is serious. Although the domestic localization rate has improved, “bottleneck” areas such as lithography machines and high-end materials still heavily depend on imports. At the same time, the self-sufficiency rate of temporary bonding adhesives used for advanced packaging is insufficient, at less than 10%. Third, intensified market competition leads to an inward-spiral crisis. As capital withdraws and inventory is cleared, the industry enters the “elimination race” stage. If core technology barriers cannot be built, companies are highly likely to fall into low-end internal friction.

A research report from Oriental Fortune Securities released three risk warnings. First, macro factors such as international situations affect demand recovery. Uncertainty in global politics and economics—such as trade disputes, geopolitical tensions, and economic sanctions—may negatively impact market demand. These macro factors may lead to reduced consumer confidence and lower business investment, thereby suppressing demand recovery.

Second, key technology breakthroughs may not meet expectations. Technological progress is a key factor driving industry development. If R&D progress for key technologies is slow or fails to achieve expected breakthroughs, it may hinder product innovation and upgrades, affecting a company’s competitiveness and market performance.

Third, intensified China-U.S. friction leads to global industrial chain restructuring. Trade friction between China and the U.S. may force companies to reconsider their supply-chain layout to avoid potential trade barriers and tariffs. Such restructuring may increase costs, reduce production efficiency, and even affect the stability of global supply chains.

A research analyst at West China Securities said that at the current point in time, attention should be paid to the disturbance caused by overseas uncertainties to the global AI investment rhythm. However, since the domestic market has relatively resilient demand, domestic models’ cost/ecosystem provide strong hedging, and the entire domestic computing-power supply chain is expected to continue to improve steadily under a dual drive of “policy + cost.”