The rapid iteration of large AI models, the large-scale construction of intelligent computing centers, and the comprehensive penetration of edge intelligent scenarios continue to drive up the demand for domestic AI computing power. Domestic computing power chip companies are entering a critical period of performance realization.
By April 30, 2026, Cambricon Technologies Corporation Limited, Hygon Information Technology Co.,Ltd., Moore Threads, and MetaX have all disclosed their financial reports for the first quarter of 2026. Biren Technology and Iluvatar Corex, two companies listed in Hong Kong, have not yet released their first-quarter data. Overall, the industry shows a pattern where leading firms maintain stable profitability, second-tier manufacturers are accelerating their pace of loss reduction or even turning a profit, and a tiered growth structure is becoming clear.
Among them, three companies achieved profitability. Both Cambricon Technologies Corporation Limited and Moore Threads saw year-on-year revenue growth exceeding 150%. Hygon Information Technology Co.,Ltd. reported first-quarter revenue surpassing 40 billion yuan, further validating the industry's sustained high growth. Domestic AI computing chips are transitioning from technological breakthroughs to large-scale commercial implementation, with the localization process continuously deepening.
As of the market close on April 30, 2026, among the six major domestic AI computing power chip companies, the leading A-share firms lead in scale: Cambricon Technologies Corporation Limited had a market capitalization of 716.848 billion yuan, and Hygon Information Technology Co.,Ltd. followed with a market cap of 688.585 billion yuan, ranking first and second in the industry. Moore Threads and MetaX followed closely, with latest market capitalizations of 336.399 billion yuan and 303.288 billion yuan, respectively. The Hong Kong-listed Iluvatar Corex and Biren Technology had relatively smaller market caps, approximately 103.750 billion yuan and 99.913 billion yuan when converted to RMB.
Hygon Information Technology Co.,Ltd. and Cambricon Technologies Corporation Limited both achieved new performance breakthroughs. As the dual leaders in domestic computing chips, both companies' first-quarter results firmly placed them in the top tier, with simultaneous improvements in profit quality and operational efficiency.
Cambricon Technologies Corporation Limited achieved leapfrog growth in the first quarter. The company's report showed revenue reached 28.85 billion yuan during the period, a significant year-on-year increase of 159.56%. Net profit attributable to shareholders was 10.13 billion yuan, up 185.04% year-on-year. After deducting non-recurring gains and losses, net profit was 9.34 billion yuan, surging 238.56% compared to the same period last year.
Crucially, Cambricon's operating cash flow turned positive to 8.34 billion yuan in the first quarter, an improvement of over 22 billion yuan year-on-year. This marks the first time the company has reported positive quarterly cash flow since its listing, indicating a breakthrough in commercialization and collection capabilities.
Cambricon stated that its products are being deployed in key industry scenarios such as telecommunications operators, finance, and the internet. The usability of its software platform, the stability of large-scale commercial deployment, and the versatility of its AI application scenarios have all been validated in demanding customer environments. Its products are now deployed at scale by large model algorithm companies, server manufacturers, and AI application firms, supporting the intelligent upgrade processes in cloud computing, energy, education, finance, telecommunications, healthcare, and the internet.
Financial data shows that Cambricon's first-quarter orders were also effectively supported by market demand. The company's contract liabilities at the end of the first quarter amounted to 396 million yuan, an increase of over 300 million yuan from the beginning of the quarter. Furthermore, both accounts receivable and prepayments saw substantial growth compared to the end of 2025, indicating positive signals for the company's orderly development.
Hygon Information Technology Co.,Ltd. continued its steady, high-growth trajectory, with first-quarter revenue reaching 40.34 billion yuan, a 68.06% year-on-year increase. Net profit attributable to shareholders was 6.87 billion yuan, up 35.82% year-on-year. After adjusting for non-recurring items, net profit was 5.97 billion yuan, an increase of 34.99%. During the same period, its R&D investment reached 12.11 billion yuan, a 26.8% year-on-year increase, supporting continuous technological iteration through high-intensity investment.
Hygon adheres to a dual-driver strategy of "CPU + DCU". Its DCU product, Deep Compute 3, has completed adaptation with 365 mainstream large models, including DeepSeek and Qwen3, covering 99% of the world's non-closed-source large models. It empowers a full range of scenarios from billions-of-parameters edge inference to trillion-parameter model training. The company is also actively developing its next-generation Deep Compute 4, with progress reported as smooth.
In terms of ecosystem adaptation, Hygon leverages the advantages of both CPU and DCU ecosystems. Through the光合组织 (Guanghe Organization), it has gathered over 6,000 ecosystem partners. It continues to enhance partner-focused initiatives like the "Star Sea Plan" and has launched the "Strong Core Solid Base" plan for fundamental software ecology, comprehensively connecting the ecosystem chain from core components and complete machine systems to application software, accelerating deep hardware-software synergy.
Sha Chaoqun, General Manager of Hygon Information Technology Co.,Ltd., stated at a recent earnings conference that over the next three to five years, competition in intelligent computing centers will gradually shift from comparing construction scale to comparing energy efficiency ratios, cluster utilization rates, and full lifecycle comprehensive costs.
Sha indicated that Hygon's DCU, as a domestic high-end AI acceleration computing product, is one of the few GPGPU chips in China capable of supporting both full-precision and half-precision AI training. With the growth of AI Agent and AI inference demands, scenarios involving task scheduling, logical reasoning, and multi-service collaboration significantly increase the need for general-purpose computing. Computational tasks are trending towards highly concurrent, chain-amplifying workloads, naturally raising requirements for CPU concurrency and memory capacity. "TrendForce expects the CPU-to-GPU ratio to increase from traditional levels, and the company's CPUs are expected to顺势提升 (seize the opportunity to increase) their penetration rate in the commercial market."
Moore Threads and MetaX continued to improve their product portfolios. As core players in the GPU sector, both companies accelerated their commercialization processes in the first quarter, achieving a turnaround to profitability and a significant reduction in losses, respectively, demonstrating strong growth momentum.
Moore Threads became the first listed domestic GPU company to achieve quarterly profitability. In the first quarter of this year, the company reported revenue of 7.38 billion yuan, a year-on-year increase of 155.35%. Net profit attributable to shareholders was 29.36 million yuan, turning a profit compared to a loss of 112 million yuan in the same period last year. After adjusting for non-recurring items, the net loss was 54 million yuan, with the loss narrowing by 60.1%.
Regarding performance growth, Moore Threads stated that the AI explosion has rapidly increased market demand for GPUs. In 2025, rapid iteration of large models by major manufacturers, the blossoming of AI applications, and the rapid development of emerging fields like embodied AI and autonomous driving significantly boosted demand for AI chips, represented by GPUs. On one hand, restricted by US export policies on high-end GPUs, domestic AI chips are gradually replacing international GPU products, presenting a historic development opportunity. On the other hand, the company's full-feature GPU, known for its strong versatility and support for all computing precisions, provides good support for current multimodal and converged computing scenarios.
In R&D, Moore Threads invested 3.69 billion yuan in the first quarter, a 50% year-on-year increase, continuing to bolster product iteration.
During the first quarter, Moore Threads secured a major 6.6 billion yuan order for an intelligent computing cluster, achieving large-scale deployment of thousand-card/ten-thousand-card clusters. Its "Kua E" series chips were deployed in batches and launched services in intelligent computing centers and by cloud providers. The new "Hua Gang" architecture is advancing the construction of hundred-thousand-card clusters, with efforts focused on both consumer-grade and data center GPUs.
Regarding future product development plans, Zhang Jianzhong, Chairman and General Manager of Moore Threads, stated at a recent earnings conference that the company will launch the high-performance AI training-and-inference integrated "Hua Shan" chip and the "Lu Shan" chip specialized for high-performance graphics rendering, both based on the "Hua Gang" architecture. "Hua Shan" focuses on integrated AI training/inference and ultra-large-scale intelligent computing. It integrates a new generation of asynchronous programming and full-precision tensor computing units, supporting computation from FP4 to FP64, providing stable and efficient computing power support for ten-thousand-card and larger intelligent computing clusters, serving as a solid foundation for building the next generation of "AI factories." "Lu Shan" specializes in high-performance graphics rendering, achieving comprehensive performance leaps, including a 16x improvement in geometry processing performance, a 64x increase in AI computing performance, a 50x boost in ray tracing performance, alongside significantly enhanced texture filling, atomic memory access capabilities, and video memory capacity.
MetaX reported high revenue growth and a significant narrowing of losses in the first quarter. During the reporting period, it achieved revenue of 5.62 billion yuan, a 75.37% year-on-year increase. The net loss attributable to shareholders was 99 million yuan, a 57.49% reduction compared to the loss in the same period last year. R&D investment was 253 million yuan, primarily focused on enhancing the product portfolio.
To date, MetaX has established three core product lines: Xi Yun C (focusing on training), Xi Si N (focusing on inference), and Xi Cai G (focusing on graphics applications). In the first quarter, it launched the Xi Suo X series scientific intelligence chip, entering high-end scenarios like climate simulation, fluid dynamics, and molecular simulation. Customer expansion and product shipments are expected to increase significantly.
Regarding the company's 2026 development plan, MetaX stated at the earnings conference that looking ahead, the high growth of the AI chip industry and the acceleration of domestic substitution will broaden the market space. The company's products have promising market prospects and vast potential. Revenue growth is expected to remain high. With ongoing cost optimization and expense control, gross margin and period expense ratios are expected to stabilize, and the volume sales of new products will continuously contribute to performance.
Biren Technology and Iluvatar Corex are positioning themselves in the high-end computing power segment and have delivered multiple computing cluster projects. While they have not yet disclosed their Q1 2026 financial reports, operational progress continues, focusing on securing a position in the high-end computing power race.
Biren Technology, a representative of domestic high-end AI training chips, reported full-year 2025 revenue of 10.3 billion yuan, a 207% increase over the previous year. Its adjusted annual loss was 8.7 billion yuan, compared to a loss of 7.7 billion yuan in the prior year. In 2025, its gross profit was 5.57 billion yuan, surging 210.8% year-on-year, with a gross margin reaching 53.8%, an increase of 63 basis points (bps).
Biren Technology stated that it successfully delivered multiple large-scale intelligent computing cluster projects, including a 2048-card optical interconnect and optical switching GPU super-node cluster and several market-oriented thousand-card level intelligent computing clusters. Its customers encompass national-level computing power platforms, telecom operators, commercial AIDCs, AI/large model companies, and enterprise clients.
Iluvatar Corex focuses on the general-purpose computing chip track. In 2025, the company achieved revenue of 10.34 billion yuan, a 91.6% year-on-year increase. Gross profit was 5.58 billion yuan, surging 110.5% year-on-year, with gross profit growth outpacing revenue growth. During the same period, its R&D and operational efficiency improved significantly, with the adjusted net loss narrowing by 32.1% year-on-year.
In specific business terms, high-speed breakthroughs in both the general-purpose GPU training and inference tracks became the core support for its performance growth. In 2025, Iluvatar Corex's general-purpose GPU business revenue reached 9.23 billion yuan, a 149.6% year-on-year increase, accounting for 89.3% of total revenue, highlighting strong business focus and growth potential.
By the end of 2025, Iluvatar Corex had cumulatively served over 340 industry customers, with products and solutions deployed in more than 1,000 projects. It was among the first to achieve widespread coverage in core areas such as the internet, AI large models, scientific research, finance, healthcare, education, and transportation. Its customer base continues to expand, and scale effects are gradually emerging.
It is noteworthy that high-intensity R&D investment remains the core method for companies to build technical moats and navigate industry cycles.
In the first quarter, the R&D expenditure-to-revenue ratio for the four A-share computing chip companies that disclosed reports remained high. Among them, Hygon Information Technology Co.,Ltd.'s R&D investment was 8.6 billion yuan. The R&D expenditure ratios for both Moore Threads and MetaX exceeded 45%, indicating continued heavy investment in core technology攻坚 (breakthroughs).
The industry generally believes that the AI computing chip sector features fast technological iteration and high R&D barriers. Only sustained R&D investment can achieve technological breakthroughs in high-end segments and build ecosystem barriers.
Tang Yue, an analyst at Zhongyuan Securities, stated in a research report released in April 2026 that the development speed of AI applications in 2026 overall exceeded expectations, leading to tightness across the computing power industry supply chain. For 2026, against the backdrop of constrained supply of overseas chips, domestic AI chips have also seen improvements in both performance and capacity, and are expected to continue benefiting from changes in market structure, ultimately improving the supply capacity of domestic computing power.
Comments