Domestic listed chip manufacturers have recently seen weaker stock performance, influenced by factors such as major tech companies developing their own chips and limited domestic advanced process capacity. Institutions believe that domestic AI models have a strong demand for computing power, and the performance of domestic chip companies is expected to further improve.
Recently, domestic AI chip companies have successively released their 2025 annual results, showing widespread explosive growth in performance.
On the evening of March 12, Cambricon Technologies Corporation Limited (688256.SH) was the first to release its 2025 financial report. The report showed that Cambricon achieved operating revenue of 64.97 billion yuan in 2025, a significant increase of 453.21% year-on-year. It reported a net profit attributable to shareholders of 2.059 billion yuan, turning a profit compared to a loss of 452.3 million yuan in the same period last year. This marks Cambricon's first annual profit since its establishment in 2016.
On March 13, the first trading day after the earnings release, Cambricon's stock price opened slightly higher but later fell back. By the close, the stock was down 0.26% to 1,096.10 yuan per share.
According to the financial report, after turning a profit, Cambricon will conduct its first dividend distribution, proposing a cash dividend of 15.00 yuan per 10 shares (before tax) for all shareholders. Additionally, the company plans to transfer 4.9 shares per 10 shares from its capital reserve. After the stock dividend, Cambricon's per-share price will decrease significantly. In August 2025, Cambricon's stock price once surpassed that of Kweichow Moutai (600519.SH) to become the highest-priced stock on the A-share market, drawing significant market attention.
Two other chip companies, Moore Threads and MetaX, which listed at the end of 2025, have also recently released their 2025 performance forecasts, showing a substantial narrowing of losses. Moore Threads' operating revenue increased 2.43 times year-on-year to 1.506 billion yuan, with its net loss narrowing by 36.70% to 1.024 billion yuan. MetaX's operating revenue grew 1.21 times year-on-year to 1.644 billion yuan, and its net loss narrowed by 44.53% to 781 million yuan.
The accelerated path to profitability for domestic AI chip companies is closely linked to the surge in computing power demand driven by AI development and the accelerated trend of domestic chip substitution. Their revenues have shown rapid growth in recent years. From 2022 to 2025, Cambricon's revenue grew from 729 million yuan to 64.97 billion yuan, an increase of 7.9 times. Moore Threads' revenue increased from 46 million yuan to 1.506 billion yuan, a 31.7-fold rise. MetaX's revenue surged from 426,400 yuan to 16.44 billion yuan, an astonishing increase of 38,545 times.
Especially after listing, domestic AI chip companies have entered a new phase of development. The GPU industry is characterized by high technical barriers, significant R&D investment, and long development cycles. Support from capital markets not only provides ample R&D funding but also enhances brand influence and attractiveness to top talent, enabling stronger commitment to independent innovation in domestic GPUs.
However, domestic AI chip stocks have performed weakly this year, with Cambricon, Moore Threads, and MetaX all recording declines year-to-date.
The recent weak performance of domestic computing power stocks like Cambricon is attributed to earlier rumors during the earnings quiet period, as well as factors such as the approval of NVIDIA's H200 exports to China, major companies developing their own chips, and limited domestic advanced process capacity, which have impacted optimistic expectations for future performance. Particularly regarding capacity issues, Cambricon faced certain delivery challenges in 2025. However, the demand curve for domestic computing power is steep this year, with domestic AI models showing strong demand. Delivery issues are expected to be gradually resolved, and performance is anticipated to improve further.
GF Securities stated that Cambricon's revenue saw significant growth in 2025, with notable commercial expansion in the internet sector. It is expected that sales of the company's AI chips to commercial clients like internet companies will reach a new level in 2026. The substantial increase in inventory in the fourth quarter reflects the company's optimistic expectations for future product deliveries and enhances the sustainability and certainty of high revenue growth in subsequent quarters. Compared to the uncertainty surrounding NVIDIA's H200 chip sales in China, the performance improvements and supply chain recovery of domestic AI chips provide more stable computing power support for domestic tech companies. As the company's position in the industry chain and technological competitiveness gain market consensus, its long-term premium is expected to become more prominent.
Zhongtai Securities indicated that achieving independent and controllable computing power is a definite direction. The development of domestic computing power will focus on achieving self-sufficiency, reducing reliance on external technology, and enhancing the competitiveness of the domestic AI chip industry. As AI computing power demand increases, domestic chips will gradually evolve from single cards to system-level integration, improving overall performance and energy efficiency. 2026 will be a critical year for the scaling development of domestic computing power.
Looking at the released 2025 results of Cambricon, Moore Threads, and MetaX, all three domestic AI chip companies showed substantial revenue growth and either achieved profitability or accelerated their loss reduction.
As a leading domestic computing power company, Cambricon achieved its first annual profit since 2016. According to the financial report, Cambricon's operating revenue in 2025 was 64.97 billion yuan, a year-on-year increase of 453.21%. Gross profit totaled 35.83 billion yuan, up 437.99% year-on-year. Along with significant revenue growth, the company achieved its first annual profit, with net profit attributable to shareholders reaching 2.059 billion yuan and adjusted net profit after non-recurring items at 1.77 billion yuan.
Cambricon stated that benefiting from the continuous rise in computing power demand in the AI industry, the company expanded its market presence with its competitive products and actively promoted the implementation of AI application scenarios, leading to a substantial year-on-year increase in revenue scale for 2025.
According to the financial data, Cambricon's profitability improved significantly. Its gross profit margin for 2025 was 55.15%, and its net profit margin was 31.68%. While the gross margin slightly decreased from 56.71% in 2024, the net profit margin improved dramatically from -38.91% in 2024. R&D expenses in 2025 were 1.351 billion yuan, an increase of 11.1% year-on-year, but the R&D expense ratio dropped significantly from 103.57% in 2024 to 20.79% in 2025. The return on equity reached 26.96%. However, there is still a gap compared to NVIDIA's net profit margin of 55.6% and return on equity of 101.49% for its fiscal year 2026.
The financial report also showed that in 2025, Cambricon produced and sold 128,000 and 117,400 smart chips and board cards, respectively. Inventory value at the end of 2025 reached 4.9 billion yuan, an increase of approximately 1.2 billion yuan from the third quarter of 2025. According to the annual report, the inventory of smart chips and board cards was 857,000 units, which will support further product deliveries to customers.
After turning a profit, Cambricon will conduct its first dividend distribution, proposing a cash dividend of 15.00 yuan per 10 shares (before tax) for all shareholders, totaling 632 million yuan (before tax), accounting for 30.71% of the 2025 net profit attributable to shareholders. Additionally, the company plans a capital reserve transfer of 4.9 shares per 10 shares.
The report also indicated that well-known investor Zhang Jianping further increased his holdings in Cambricon, raising his position by 408,400 shares to 6.8149 million shares, representing 1.62% of the circulating shares and making him the fifth-largest shareholder. Based on the closing price on March 13, Zhang's holdings are valued at nearly 7.5 billion yuan. This marks the second consecutive quarter of increased holdings by Zhang, who had added 320,200 shares in the third quarter of 2025.
Notably, with the achievement of annual profitability in 2025, Cambricon successfully removed the "U" suffix and exited the STAR Market Growth Layer on March 16. The stock abbreviation changed from Cambricon-U to Cambricon. The STAR Market Growth Layer was established in July 2025 specifically for unprofitable tech companies, with a grandfathering rule allowing existing companies to exit the layer upon their first profit, while new registrants must meet higher profitability standards. Cambricon's exit means it will be among the first batch of companies to leave the Growth Layer.
Meanwhile, Moore Threads and MetaX, two leading domestic GPU companies that listed on the STAR Market in December 2025, also achieved rapid revenue growth and significant loss reduction.
According to previously disclosed performance forecasts, Moore Threads achieved operating revenue of 1.506 billion yuan in 2025, a year-on-year increase of 2.43 times. The net loss attributable to shareholders was 1.024 billion yuan, a substantial reduction of 594 million yuan compared to the loss of 1.618 billion yuan in 2024, narrowing the loss by 36.70%.
Moore Threads stated that benefiting from the development of the AI industry and strong demand for high-performance GPUs, the company's product competitiveness and market recognition continued to improve, driving revenue and gross profit growth and narrowing the loss year-on-year. The company also noted that it is still in a high R&D investment phase and, compared to international industry giants, there are gaps in comprehensive R&D strength, core technology accumulation, and product ecosystem. It is currently not profitable and has accumulated losses.
Another listed domestic GPU company, MetaX, reported operating revenue of 1.644 billion yuan in 2025, a year-on-year increase of 1.21 times. The net loss attributable to shareholders was 781 million yuan, a significant reduction of 628 million yuan from the loss of 1.409 billion yuan in 2024, narrowing the loss by 44.53%. This marks the first year of reduced losses for MetaX since 2022.
MetaX stated that in 2025, it adhered to its '1+6+X' development strategy, intensified market expansion efforts, continuously enhanced its market position and influence in the high-performance GPU industry, and promoted the deep integration of AI technology across various sectors. The company's GPU product shipments increased significantly, driving a notable year-on-year increase in revenue scale. Additionally, the substantial year-on-year growth in operating revenue and a reduction in share-based payment expenses positively impacted the company's 2025 profit, reducing the loss and showing an improving trend in operating performance.
MetaX also took the lead in releasing its first-quarter 2026 performance forecast, indicating a further narrowing of losses. The company expects revenue of 400 million to 600 million yuan, a year-on-year increase of 24.84% to 87.26%. The estimated net loss attributable to shareholders is 91 million to 182 million yuan, a narrowing of 21.93% to 60.97%.
In the first-quarter forecast, MetaX mentioned that benefiting from the rapid development of the AI industry and relying on excellent product performance and a complete software ecosystem, the company's products and services have gained widespread recognition from downstream customers, leading to significant year-on-year business growth.
It is worth noting that although Moore Threads and MetaX have not yet turned a profit in 2025 and remain in a loss position, both companies have provided forecasts for their profitability timelines.
MetaX stated that based on its own operating conditions, combined with product market space, market share changes, customer repurchases, and new customer verification and expansion, it expects to reach the break-even point as early as 2026.
Moore Threads predicts that it can achieve consolidated profitability in 2027, which includes profits from government subsidies. Excluding these subsidies, the company would be marginally profitable in 2027.
Currently, two Hong Kong-listed domestic GPU companies have not yet announced their 2025 results. However, according to their interim reports for 2025, both showed a year-on-year expansion in net losses.
Biren Technology reported operating revenue of 58.9 million yuan for the first half of 2025, a nearly 50% year-on-year increase, but its net loss attributable to shareholders was 1.601 billion yuan, an increase of 80.17% year-on-year. Iluvatar Corex reported revenue of 324.3 million yuan for the first half of 2025, a 64.24% year-on-year increase, but its net loss attributable to shareholders was 609.3 million yuan, an increase of 50.82% year-on-year.
As a leading domestic AI chip company, Cambricon was the first in the industry to break through the breakeven point, achieving annual profitability in 2025. This was primarily due to its domestic ASIC chips for AI data centers gaining recognition from major domestic cloud providers, entering their supply chains and securing large orders. The cost dilution effect contributed significantly to turning a profit.
According to the financial report, Cambricon's operating revenue began explosive growth starting in the third quarter of 2024. In Q3 2024, Cambricon achieved quarterly revenue of 120.5 million yuan, a substantial increase of 284.59% year-on-year. Revenue further grew by 75.51% in Q4 to 989.2 million yuan, and the company achieved its first quarterly net profit of 272.2 million yuan.
According to its 2024 annual report, sales to the largest customer reached 930 million yuan, nearly doubling from 467 million yuan the previous year, a significant increase of 99.1%. Combined sales to the second to fifth largest customers were 182 million yuan, slightly lower than the previous year's 188 million yuan, indicating the dominant role of the top customer in driving revenue.
Notably, in 2025, Cambricon's customer concentration decreased. From 2023 to 2024, sales to the top five customers accounted for 92.36% and 94.63% of operating revenue, respectively. In 2025, sales to the top five customers were 5.76 billion yuan, accounting for 88.66% of operating revenue. Among the top five customers in 2025, the third customer was a long-term partner, while the others were new customers added during the period. Sales to the first, second, fourth, and fifth customers were 1.703 billion yuan, 1.401 billion yuan, 764 million yuan, and 655 million yuan, respectively.
This indicates that Cambricon expanded its customer base in 2025, adding four new customers among its top five. The sales contributions from the new first, second, fourth, and fifth customers were 26.21%, 21.56%, 11.76%, and 10.08%, respectively. The share of the largest customer decreased significantly from 79.15% in 2024 to 26.21%.
In late February, industry sources indicated that Cambricon's 590 chip entered Tencent's supply chain. Previously, in September 2025, there were reports that Alibaba increased its orders for Cambricon's Siyuan 370 chips to 150,000 units. However, these reports have not been confirmed.
Cambricon mentioned in its 2025 financial report that its products now widely serve large model algorithm companies, server manufacturers, and AI application companies, impacting intelligent upgrades in sectors like cloud computing, energy, education, finance, telecommunications, healthcare, and internet, supporting the rapid development of the AI industry. In the internet sector, focusing on core application scenarios like large models and multimodality, the company has deepened technical cooperation with various industry customers in operator development and optimization, framework optimization, and communication optimization.
An insider from a domestic chip company explained that domestic AI chip companies are mainly divided into those affiliated with major tech firms and independent startups, each with its own advantages and disadvantages.
Regarding channels, when product performance is similar, chip companies affiliated with major tech firms undoubtedly have an advantage in procurement, as these firms prefer to support their own chip companies. However, such companies may have weaker customer acquisition capabilities in other sectors. Independent AI computing power companies face higher barriers to entering the procurement systems of major tech firms but have stronger capabilities in expanding industry clients outside these firms. Encouragingly, independent AI chips now have the potential to enter major tech procurement systems, but they must undergo performance testing and comparison. If their cost-performance ratio stands out, their products are likely to be adopted, as procurement primarily focuses on cost-effectiveness.
Notably, looking at the stock performance of the five listed domestic AI chip companies, most have performed poorly this year.
Moore Threads and MetaX, which listed on the STAR Market at the end of 2025, saw enthusiastic performance initially but later experienced sluggish stock prices. As of March 12, the stock prices of these two domestic GPU companies had fallen 40.79% and 41.9%, respectively, from their highs.
Cambricon, which listed earlier, saw significant stock price increases from 2023 to 2025. However, entering 2026, the upward trend slowed. Although Cambricon's 2025 report showed a turnaround to profitability, its stock price has been declining since hitting a high of 1,550 yuan per share on January 12, 2026. By March 12, the stock had retreated 29.1%, with its total market capitalization dropping from 654.1 billion yuan to 463.43 billion yuan, a decrease of 190.67 billion yuan.
The performance of the two domestic GPU companies listed in Hong Kong this year has been mixed. Biren Technology's stock price continued to decline after reaching a high of 42.88 Hong Kong dollars per share, falling to 33.22 Hong Kong dollars by March 12, a drop of 22.53% from the high. Iluvatar Corex's stock performance has been relatively strong, continuously reaching new highs after listing.
Why have domestic AI chip companies seen accelerated profit improvement but poor stock performance? Market analysis suggests several reasons. First, high valuations at the time of listing affected Moore Threads and MetaX, which listed at the end of 2025. Due to the scarcity of such targets and benefits from the global AI wave and domestic substitution trends, coupled with substantial returns for new shares and a relatively small floating supply initially, they attracted speculative capital. Their stock prices rose to highs of 941.08 yuan and 895 yuan per share, respectively, representing increases of 7.23 times and 7.55 times over their issue prices of 114.28 yuan and 104.66 yuan per share. Based on price-to-sales ratios calculated for the third quarter of 2025, they were as high as 468 times and 189 times, far exceeding Cambricon's approximately 108 times P/S ratio during the same period.
Consequently, as speculative capital receded and lock-up shares faced gradual release, both stocks experienced valuation corrections. As of March 11, the P/S ratios for Moore Threads and MetaX were 285 times and 122 times, respectively.
Second, as more domestic AI chip companies list on A-share and Hong Kong capital markets, reduced scarcity has led the market to scrutinize the fundamental quality of each company more carefully, focusing on future performance trends and industry competition.
Third, competition in the domestic AI chip market is intensifying. Cloud provider customers of domestic AI chip companies are also accelerating the development of their own AI chips. According to previous reports, Alibaba's affiliated AI chip company, T-Head, has shipped hundreds of thousands of its Zhenwu PPU chips, surpassing Cambricon and leading among domestic GPU manufacturers, though this information has not been further confirmed by Cambricon. Baidu's affiliated Kunlun Chip secretly submitted a main board listing application to the Hong Kong Stock Exchange on January 1. Its chips are used not only for Baidu's internal AI projects but are also supplied to other industry clients.
More notably, it is reported that ByteDance, previously speculated to be a major customer of Cambricon, is developing an AI chip codenamed SeedChip. It has contacted Samsung Electronics for manufacturing and plans to receive samples by the end of March, aiming to produce at least 100,000 chips for AI inference in 2026 and gradually increase capacity to 350,000 units.
Additionally, as the biggest common threat to domestic chips, NVIDIA's H200 chip was approved for export to China by the Trump administration on December 8, 2025, marking the first "limited opening" for high-end NVIDIA AI chips since the comprehensive ban from 2022 to 2025. However, on February 24, David Peters, Assistant Secretary of Commerce for Export Enforcement, confirmed at a congressional hearing that although the Trump administration approved NVIDIA's export of H200 chips to China, no sales have been recorded to Chinese clients thus far. Despite this, for domestic chip manufacturers, this remains a factor affecting investor confidence and optimistic future performance expectations.
Beyond market competition, domestic AI chips generally face challenges related to insufficient domestic advanced process capacity, leading to relatively slow deliveries, which constrains performance growth. The biggest issue currently facing the domestic AI chip industry is the supply chain problem, specifically capacity constraints. Foreign chip manufacturers can sustainably supply high-end chips using 3nm processes or more advanced technologies, while the most advanced process achievable for large-scale production in mainland China is the equivalent of 7nm, and output is limited.
Oriental Securities believes that China's model invocation volume has surpassed that of the US, leading to an explosion in inference demand. The continued expansion of domestic computing power demand is directly driving growth in computing chip demand. Against the backdrop of restricted supply of high-end overseas computing power, domestic computing power is achieving breakthroughs in computing density, ecosystem adaptation, and commercial realization, gradually enabling scaled substitution.
Guotou Securities stated that driven by both external pressure and internal demand, domestic computing power is accelerating its transition from a "usable" alternative to a "high-performing" mainstream solution. With domestic cloud providers' capital expenditures entering an upward trajectory, the widespread application of lightweight models, and the continued maturation of the domestic computing power ecosystem, domestic computing infrastructure is expected to achieve scaled deployment and value realization in key industries such as government affairs, finance, internet, and smart manufacturing.
Comments