"Hong Kong Stock Exchange's GPU First Stock" Annual Report Submitted: Losing More, Earning More?

This is Biren Technology’s first annual report after its listing on the Hong Kong Stock Exchange.

On January 2, 2026, as the echoes of the New Year bell faded, Biren Technology officially rang the bell to list on the Hong Kong Stock Exchange, becoming the first GPU-related target in the Hong Kong market, and firing the first shot of the year’s hard-tech IPO boom.

Recently, the company formally released its 2025 full-year performance announcement—coinciding with key milestones such as the large-scale implementation of the global artificial intelligence industry and the continued release of demand for domestic computing power substitution. Backed by commercial breakthroughs in its core products, Biren Technology delivered rapid growth in key metrics such as revenue: during 2025, the company’s revenue first surpassed the 1 billion yuan mark, reaching RMB 1.03B, up 207.2% year over year. Gross profit also rose in tandem to RMB 557 million, up 210.8% year over year. The gross margin remained stable at 53.8%. Revenue from its core intelligent computing solutions contributed over 90%, and solid business fundamentals brought substantial performance growth.

However, behind the impressive growth curve, the company’s loss amount expanded at the same time. The full-year book loss reached RMB 16.49B, representing a year-over-year increase of as much as 972.3%. This figure was mainly affected by non-operating one-off factors such as an increase in the fair value of redemption liabilities related to investors’ redemption rights and listing expenses. The company’s redemption rights were permanently terminated after listing, and the related liabilities were also converted into equity, meaning that no such profit-and-loss impacts will arise in the future. But even after excluding the above one-off factors, the adjusted loss was still RMB 874 million, widening by 13.9% year over year.

Biren Technology’s team told Beijing Business Today that the high growth in performance during the year came from the large-scale implementation and delivery of its intelligent computing solutions business. The company is still in a rapid growth stage, prioritizing long-term capability building with strategic resources. At this stage, its focus is on R&D investment and market expansion. Relying on sufficient capital reserves, the company is also stepping up next-generation chip and system-level technology R&D to seize first-mover advantages in the computing-power market for the AI inference era.

1. The bonus period as demand surges

The full-scale explosion of large-model technology is reshaping the global intelligent computing chip market landscape, with training and inference becoming the two core application scenarios in the industry.

Early industry demand was concentrated in the pre-training phase of large models. Training and developing models with trillions of parameters—scaling to the trillion-parameter and even ten-trillion-parameter range—requires large-scale server clusters to provide massive computing power support. High bandwidth and high-speed interconnect are key requirements. As the functions of large models and application scenarios continue to mature, the frequency of terminal usage and the scope of coverage expand rapidly, driving explosive growth in computing-power demand for inference scenarios: while the computing power needed for a single task may be relatively moderate, the overall computing power consumption continues to climb due to high-concurrency characteristics, and low latency has also become a core demand for inference scenarios.

According to the prospectus filed by Biren Technology and materials cited from China Insights Consulting (CIC), over the next one to two years, the investment growth rate in inference chips is expected to surpass that of training chips, becoming the core driving force behind industry growth.

In terms of market size, the global intelligent computing chip industry continues to grow in a leapfrogging manner: from 2020 to 2024, the market size rapidly grew from $6.6 billion to $119 billion, with a compound annual growth rate (CAGR) of 106%. It is expected that by 2029 it will exceed $585 billion. From 2024 to 2029, the CAGR is expected to remain at 37.5%. China, as a global core AI market, continues to grow at a faster pace than the world average: from 2020 to 2024, market size grew from $1.7 billion to $30.1 billion, with a CAGR of 105%. It is expected that by 2029 it will reach $201.2 billion, and the CAGR from 2024 to 2029 will be 46.3%.

Among them, GPGPU (general-purpose graphics processing unit), thanks to its generality, flexibility, and a mature software ecosystem, has become the mainstream choice for intelligent computing chips. In 2024, GPGPU accounted for 92% of the global intelligent computing chip market, and 78.1% in China. The China GPGPU market is expected to have a CAGR of 49% from 2024 to 2029, indicating significant growth potential.

According to the prospectus and public information, in the context of industry competition shifting from focusing on a single chip’s performance to competing across full-stack dimensions such as hardware iteration, software ecosystems, and cluster system coordination, large-scale computing power clusters have become an industry trend. High R&D investment and ecosystem building have also become an urgent need for companies to maintain competitiveness.

2. Product execution—waiting for the flowers to bloom

Biren Technology’s breakthroughs in 2025 are reflected directly in a doubling of revenue growth. Behind the numbers lies an effective improvement in the company’s ability to implement its business.

According to the company’s introduction to Beijing Business Today, the core support for the high growth in revenue is the scaled deployment and delivery of intelligent computing solutions—its two flagship GPUs, BR106 and BR166, achieve full-volume production in all form factors. The company has also successfully upgraded from being a single-chip supplier to becoming an overall solutions provider capable of delivering intelligence-computing clusters at the thousand-card level. Benchmark projects such as the 2048-card optical interconnect super-node cluster have been smoothly implemented, and its customer base covers national-level computing power platforms, telecommunications operators, and leading AI enterprises.

From a horizontal comparison, Moore Threads, a homegrown GPU leader listed on the STAR Market, has also delivered impressive results in the same period: according to its 2025 annual performance quick report announcement, the company’s annual revenue grew 243.37% year over year, while the loss narrowed by 36.70% year over year. The two leading companies follow different rhythms in narrowing losses and uncovering profitability potential, but the mutually reinforcing high revenue growth at both firms also validates the industry’s strong momentum.

As a technology company in a phase of rapid expansion, Biren Technology’s R&D and operating expenditures will inevitably rise in parallel. In 2025, the company’s R&D spending reached RMB 1.48B, up 78.5% year over year. It is reported that the relevant funds mainly went toward the R&D of next-generation BR20X-series chips, full-stack software optimization, and optical interconnect system technology, to specifically adapt to computing-power demand in the AI inference era. Sales and marketing expenses were RMB 57.5 million, up 11.6% year over year, used to expand the sales team and develop downstream markets. General and administrative expenses rose to RMB 332 million, up 35.8% year over year. The core drivers were increases in labor and operating costs brought by the expansion of the team size, which are typical operating expenditures for a company in its growth stage.

On the ecosystem adaptation front, Biren Technology completed rapid integration with mainstream large models such as DeepSeek, Zhipu GLM, Alibaba Tongyi Qianwen, and Tencent Hunyuan. It has connected the cooperative chain among chips, large models, and cloud platforms, significantly lowering the implementation threshold for domestic computing power. In today’s industry environment where competition is shifting toward system-level efficiency, the introduction of its optical interconnect super-node solution has precisely captured a key moment in the industry’s shift from competing on single-point performance to competing across the overall system.

Wang Peng, an associate researcher at the Beijing Academy of Social Sciences, said in an interview with Beijing Business Today that domestic GPU companies are at a critical junction in development. Enterprises need to offset the rigid investment in R&D through scaled business, build a mature software ecosystem to reduce customers’ switching costs, and gradually replace single hardware sales with integrated computing-power services.

Wang Peng believes that real commercial demand is already the core support for market growth. Cluster coordination across the industrial chain and scale effects will directly determine whether a company can absorb high R&D spending and achieve sustainable profitability.

Beijing Business Today’s reporter: Tao Feng, Wang Tianyi