Author | ZeR0 Editor | Moying
XinDongXi reported on November 10 that Denglin Technology, a Suzhou-based general-purpose GPU (GPGPU) company in Jiangsu, has officially listed on the Hong Kong Stock Exchange. The offer price for Denglin Technology was HKD 231.00 per share (approximately CNY 196.50). At today's open, the stock fell by 0.87% to HKD 229.00 per share (approximately CNY 194.80). By 9:45 AM, its share price had dropped as much as 5.97% to HKD 217.20 per share (approximately CNY 184.76), with a market capitalization of HKD 13.4 billion (approximately CNY 11.4 billion).
According to the prospectus, Denglin Technology was founded in November 2017 by Li Jianwen, Wang Chenhui, Wang Ping, and the late Wang Zhenyu, among others. Chief Strategy Officer Li Jianwen and Chief Innovation Officer Wang Ping are both alumni of Tsinghua University. Leveraging its proprietary GPU+ (Minsky) architecture, the company targets Enterprise AI as its core market while striving to expand into public cloud markets and innovative fields such as robotics. Based on data from Frost & Sullivan, ranked by revenue in 2025, Denglin Technology is in the top ten of China's GPGPU market, with a market share of less than 1.0%; it is the third-largest player in China's Enterprise AI GPGPU market, holding a 1.0% market share.
Li Jianwen, Wang Chenhui, Wang Ping, Denglin Consulting, Denglin Xingyi, Denglin Longgui, Denglin Wanhua, and Lingxiao Ventures constitute a group of controlling shareholders. Following the completion of the global offering, Denglin Technology's equity structure is as follows: Denglin Consulting is wholly owned by Li Jianping. Senior independent investors include the China Internet Investment Fund, Qualcomm China, Sino-Singapore Venture Capital, and the Integrated Circuit Growth Fund. Other industrial investors include Hikvision Fund and Haina Yuzhi under Hikvision, Zhonghe Auto, and Oriental Semiconductor.
01.Revenue Reached CNY 200 Million in Four Months This Year, Exceeding Full-Year Revenue of Last Year
In 2023, 2024, 2025, and January-April 2026, Denglin Technology's revenues were CNY 28 million, CNY 83 million, CNY 171 million, and CNY 200 million, respectively. Net profits were CNY -332 million, CNY -297 million, CNY -281 million, and CNY -16 million, respectively. R&D expenses were CNY 258 million, CNY 248 million, CNY 238 million, and CNY 65 million, respectively.
During the same periods, its adjusted net profits were CNY -278 million, CNY -224 million, CNY -204 million, and CNY -21 million, respectively.
By business category, from 2023 to the end of April 2026, over 67% of revenue came from the GPGPU business, with the remainder derived from chip design services and IP licensing.
The technical services provided by Denglin Technology primarily include chip design services. The company collaborates with customers to customize architectures, optimize performance, and adapt products to specific deployment scenarios, while also offering curated IP licensing. Its GPGPUs cover enterprise data center products and edge inference products, having been deployed in six key application scenarios: smart cities, system integration, enterprise private clouds, industrial quality inspection, education, and industry large models.
02.GPGPU Sales Exceeded 10,000 Units in Four Months; Selectively Exploring Public Cloud and Robotics Markets
Key performance specifications of Denglin Technology's main product series include:
(1) Goldwasser I Product based on Minsky I: Delivers up to 256 TOPS (INT8), 64 TFLOPS (FP16), and 16 TFLOPS (FP32) computing power, equipped with 8GB to 64GB memory, with a maximum memory bandwidth of 239 GB/s and a maximum power consumption of 55W. It is primarily used for inference tasks in smart city scenarios.
(2) Goldwasser II Product based on Minsky II: Delivers up to 512 TOPS (INT8) and 256 TFLOPS (FP16) computing power, equipped with 32GB to 128GB memory, with a maximum memory bandwidth of 192 GB/s and a maximum power consumption of 95W. It supports inference and training in smart cities, intelligent computing centers, and other scenarios.
(3) KS Series Products based on Minsky III: Deliver up to 560 TOPS (INT8) and 280 TFLOPS (FP16) computing power, equipped with 8GB to 128GB memory, with a maximum memory bandwidth of 819.2 GB/s and a maximum power consumption of 130W. They support both inference and training.
The Goldwasser series products are designed for large-scale enterprise-level AI workloads. By supporting FP32, TF32, FP16, BF16, and INT8 calculation precisions, the boards can execute multi-modal tasks such as model training, inference, image rendering, and analysis in parallel within a unified infrastructure. Total GPGPU sales increased from 4,199 units in 2023 to 5,295 units in 2024, further surging to 12,144 units in 2025, and reaching 10,344 units in the first four months of 2026. Among them, shipments of the KS series increased in the second half of 2025 and are expected to continue growing in 2026.
According to the prospectus, the general price range for GPGPUs is CNY 10,000 to CNY 200,000 per unit, varying by performance, configuration, and brand. In the first four months of 2026, the average selling price of Denglin Technology's main GPGPU products ranged between CNY 5,600 and CNY 16,200.
The company plans to launch Phase 1 and Phase 2 Minsky IV products in December 2026 and the second half of 2027, respectively. Phase 1 products will continue to use an LPDDR-based on-chip memory subsystem; Phase 2 products will integrate HBM3e or other advanced memory technologies, expected to deliver up to 2,000 TFLOPS of FP8 computing power. Target customers include large intelligent computing center operators and public cloud providers. Beyond focusing on Enterprise AI, Denglin Technology plans to selectively explore opportunities in the public cloud and robotics markets. Exploration activities in the public cloud sector include early-stage technical discussions and product roadmap development, while in the robotics market, the focus is on exploring segments such as industrial robots, service robots, and embodied intelligence robots.
03.R&D Personnel Account for Nearly 70%; Fully Proprietary Core IP Portfolio Developed
As of April 30, 2026, nearly 70% of Denglin Technology's employees are R&D personnel, most of whom have more than ten years of industry experience. Its products are built on the original GPU+ (Minsky) architecture supported by a unified Hamming software stack, which integrates GPGPU cores, AI acceleration units, and high-bandwidth on-chip memory subsystems into a single chip via heterogeneous on-die methods. Based on this architecture, Denglin Technology has developed a structured and fully proprietary core intellectual property (IP) portfolio covering a wide range of computing and acceleration functions.
Unlike IP suppliers such as Arm or Imagination Technologies, whose GPU IPs are typically designed for mobile applications and lack the performance required for high-end cloud workloads, Denglin Technology focuses on cloud-scale, high-performance applications that require large-scale architectural integration and optimization.
04.Top Five Customers Account for Over 80% of Revenue; Majority of Customers Are Private Enterprises
In 2023, 2024, 2025, and January-April 2026, revenue contributions from Denglin Technology's top five customers accounted for 98.4%, 85.1%, 83.0%, and 90.4% of total revenue, respectively. Revenue from the largest customer accounted for 37.9%, 25.7%, 29.4%, and 38.0% of total revenue during the same periods, respectively.
During the same periods, purchases from the top five suppliers accounted for 83.1%, 65.2%, 56.9%, and 76.3% of total purchases, respectively. Purchases from the largest supplier accounted for 59.5%, 31.1%, 31.2%, and 21.0% of total purchases, respectively. Several of these entities are simultaneously customers and suppliers of Denglin Technology.
During the track record period, the majority of customers for Denglin Technology's GPGPU products were private enterprises, with the rest being listed companies. To support continuous capacity expansion, Denglin Technology has placed orders to secure core wafer capacity resources equivalent to approximately 1,400 wafers, while expanding packaging, testing capabilities, and PCB assembly capacity through third-party partners. Additionally, Denglin Technology is negotiating with other potential customers, with intended contract totals amounting to approximately CNY 200 million.
05.Conclusion: Domestic GPU Companies List Successively, Transforming into Full-Stack Computing Infrastructure Providers
Over the past year, the listing process for domestic GPUs has accelerated, with GPU companies such as Moore Threads, MetaX, Denglin Technology, Tianshu Zhixin, and others successively entering the capital market. According to Denglin Technology's prospectus, Chinese general-purpose GPU companies are transitioning from hardware providers to full-stack computing infrastructure builders. At the hardware level, they extend the industrial chain around self-developed GPUs; at the software level, they ensure compatibility with operating systems and optimized algorithm frameworks; and at the service level, they provide value-added solutions tailored to diverse application scenarios.
This integrated approach lays the foundation for achieving autonomous control and scalable implementation of China's AI industry.