Denglin Technology, a Chinese developer of general-purpose graphics processors (GPUs) for artificial intelligence (AI) tasks, successfully completed its IPO in Hong Kong on November 10, raising HKD 1.41 billion. In the first four months of 2026, the company's revenue reached 200.4 million yuan, surpassing the total sales volume for the entire year of 2025.
According to a Frost & Sullivan ranking, Denglin ranked third in the corporate AI GPGPU market in China, although its market share was only 1%. The question remains whether the recently listed company can translate rapid growth into sustainable market influence.
As artificial intelligence integrates more actively into various fields—from meeting assistants to exam control systems and factory cameras—the demand for GPUs is growing. For chip manufacturers, creating a processor is only half the battle; the more complex part is winning customers from established suppliers like Nvidia.
Denglin was founded in Shanghai in 2017 by four former colleagues from Vivante, an embedded GPU developer. After six months of working with clients and suppliers, the team finalized a design that combines general computing power with acceleration for AI.
Initially, the company focused on smart city applications, and the first chip samples appeared in early 2021. As the business expanded, the headquarters was moved to Suzhou in 2024 to get closer to the semiconductor supply chain.
Denglin's early strategic choice was to use LPDDR memory (low power consumption, double speed) instead of high-performance HBM memory, which is often used in high-end accelerators. Although LPDDR is cheaper, more energy-efficient, and provides greater control over the supply chain, its lower bandwidth could become a bottleneck for the largest AI models.
To mitigate this limitation, Denglin implemented more on-chip static random-access memory (SRAM) and developed software that reuses data instead of repeatedly fetching it from external memory. The company's design prioritizes cost-effectiveness and energy efficiency over peak performance.
In October, the company demonstrated that its KS GPUs passed compatibility testing and efficiency optimization for the DeepSeek-V4.1-Flash-1017 model, confirming its ability to support a large AI model.
Denglin also plans to enter the high-performance computing market for AI. The first version of Minsky IV is scheduled for December 2026, while the second version, planned for the second half of 2027, will need to utilize HBM3e or other advanced memory types for large data centers and public clouds.
Currently, Denglin's efficiency-focused GPUs are best suited for enterprise AI, especially private clouds where clients value data control, low power consumption, and cost. The products are also aimed at edge inference and terminal devices.
Frost & Sullivan estimates that enterprise AI still represented a relatively small segment, accounting for 9.5% of the overall GPGPU market in China in 2025, compared to 90.5% in the public cloud AI segment.
Denglin's own business model has since shifted in this direction. In 2023, smart city projects accounted for about 80% of the company's GPGPU product revenue. However, by the first four months of 2026, the share of enterprise private clouds grew to 83%, while smart cities contributed only 12.2%. The KS inference cards helped facilitate this transition: during the reporting period, Denglin sold 10,246 units, generating revenue of 166 million yuan.
This shift coincided with a sharp increase in revenue. Sales grew from 27.9 million yuan in 2023 to 82.6 million yuan in 2024, then to 171.3 million yuan in 2025, and reached 200.4 million yuan in the first four months of 2026 alone.
Nevertheless, the competitive gap remains significant. According to Frost & Sullivan, in 2025, Denglin ranked third in terms of revenue in the Chinese corporate AI GPGPU market, but its share was only 1%, while Nvidia held 83.1% and AMD held 12.3%.
Rapid sales growth has not yet translated into stable profitability. In 2023, Denglin recorded a net loss of 332 million yuan, 296.5 million yuan in 2024, and 281 million yuan in 2025. The main factor was intensive investment in research and development: in 2025, it reached 238 million yuan, nearly three times the company's gross profit of 84.3 million yuan.
The situation improved in the first four months of 2026, when the net loss decreased to 16.5 million yuan compared to 108.3 million yuan the previous year. Adjusted net profit was 20.6 million yuan after accounting for stock option and listing expenses.
However, another issue relates to the customer base. The five largest customers accounted for 90.4% of the revenue in the first four months of 2026, with the largest single client contributing 38%. Converting early adoption into repeat orders and expanding the customer base will be crucial for sustaining growth.
On November 10, Denglin went public in Hong Kong, raising approximately HKD 1.415 billion. The company listed under Chapter 18C, which allows specialized technology companies that have not yet reached the exchange's annual revenue threshold of HKD 250 million to go public as preliminary issuers.
The company plans to allocate 35% of the net proceeds to upgrading existing GPU lines, 25% to developing new product categories, and 20% to software tools.
Artificial intelligence is already penetrating factories and offices. Denglin has secured its place in the market, but turning initial interest into a consistent business and long-term competitiveness presents a different challenge.