Editor’s note: Denglin Technology, a Chinese maker of general-purpose GPUs for AI computing, went public in Hong Kong on November 10, raising HK$1.41 billion. Revenue reached RMB 200.4 million in the first four months of 2026, already exceeding its full-year 2025 total. Frost & Sullivan ranked it third in China’s enterprise AI GPGPU market, but its share was just 1%. Can the newly listed chipmaker turn rapid growth into lasting market clout?
From meeting-note assistants to factory cameras and exam-monitoring systems, AI increasingly runs on GPUs. For chipmakers, building a processor is only half the job; the harder part is winning customers from established suppliers such as Nvidia.
A Different Kind of Memory Bet
Founded in Shanghai in 2017 by four former colleagues at embedded-GPU developer Vivante, Denglin spent six months visiting customers and suppliers before settling on a design combining general-purpose computing with AI acceleration.
The team initially targeted smart-city applications, with its first chip samples arriving in early 2021. As the business expanded, it moved its headquarters to Suzhou in 2024, closer to the semiconductor supply chain.
An early choice reflected Denglin’s strategy: smartphone-style low-power double data rate (LPDDR) over the high-bandwidth memory (HBM) common in high-end accelerators. LPDDR is cheaper, more energy-efficient and offers greater supply-chain control, but its lower bandwidth can bottleneck the largest AI models.
Denglin sought to ease this bottleneck with more on-chip static random-access memory (SRAM) and software that reuses data rather than repeatedly fetching it from external memory. The design prioritizes cost and power efficiency over peak performance.
In October, Denglin’s KS GPUs had completed compatibility testing and efficiency optimization for DeepSeek-V4.1-Flash-1017, demonstrating their ability to support a major AI model.
Denglin also plans to move into higher-performance AI computing. The first Minsky IV version is planned for December 2026, while a second version, targeted for the second half of 2027, is expected to adopt HBM3e or other advanced memory for large AI data centers and public clouds.
Betting on a Smaller Cloud
For now, Denglin’s efficiency-focused GPUs are best suited to enterprise AI, particularly private clouds, where customers value cost, power efficiency and data control. Its products also target edge inference and terminal devices.
Enterprise AI was still a relatively small segment, accounting for 9.5% of China’s general purpose graphics processing units (GPGPU) market in 2025, compared with 90.5% for public-cloud AI, according to Frost & Sullivan estimates.
Denglin’s own business has since shifted in that direction. In 2023, smart-city projects generated about 80% of its GPGPU product revenue.
But by the first four months of 2026, enterprise private cloud accounted for 83%, while smart cities contributed just 12.2%. Its KS inference cards helped drive the shift: Denglin sold 10,246 during the period, generating RMB166 million.
The shift coincided with rapid revenue growth. Sales climbed from RMB27.9 million in 2023 to RMB82.6 million in 2024 and RMB171.3 million in 2025, before reaching RMB200.4 million in just the first four months of 2026.
The competitive gap remains wide, however. Frost & Sullivan ranked Denglin third in China’s enterprise AI GPGPU market by 2025 revenue. Yet it held just 1% market share, compared with Nvidia’s 83.1% and AMD’s 12.3%.
Growth With a Catch
Rapid sales growth has yet to translate into sustained profitability. Denglin recorded net losses RMB332 million in 2023, RMB 296.5 million in 2024 and RMB281 million in 2025. Heavy R&D spending was a major factor: in 2025, it reached RMB238 million, nearly three times the company’s gross profit of RMB84.3 million.
The picture improved in the first four months of 2026, when its net loss fell to RMB16.5 million from RMB108.3 million a year earlier. Adjusted net profit reached RMB20.6 million after adding back share-based payment and listing expenses.
But another challenge lies in its customer base. Its five largest customers accounted for 90.4% of revenue in the first four months of 2026, with the biggest contributing 38%. Converting early adoption into repeat orders and broadening its customer base will be crucial to sustaining growth.
The Public-Market Test
On November 10, Denglin listed in Hong Kong, raising about HK$1.415 billion. It went public under Chapter 18C, which allows specialist-technology companies that have yet to meet the exchange’s HK$250 million annual revenue threshold to list as pre-commercial issuers.
The company plans to use 35% of net proceeds to upgrade existing GPU lines, 25% to develop new product categories and 20% on software tools.
AI is already moving into factories and offices. Denglin has gained a foothold, but turning early adoption into repeat business and lasting competitiveness is a different challenge.
DragonLab cuts through the noise to find the Chinese startups, founders, and technologies that could shape what comes next.
Have a startup we should know about? Reach us at dragonlab@technode.com.
