Economic Observer reporter Zheng Chenye
How many highlights can a chip have?
Probably two. The first is when it is installed in the latest smartphone, directly determining whether the phone runs fast or lags during gaming; the second is when it is installed in a massive server, determining how intelligent the AI model can become.
However, Rockchip (603893.SH) happens to be involved in neither.
According to the company’s official statement in its semi-annual report, its chips are primarily used in the AIoT field—combining artificial intelligence (AI) with the Internet of Things (IoT). Its business covers various directions including automotive electronics, machine vision, industrial applications, robotics, as well as commercial finance, education, office, and smart home.
The characteristic of this market is that customers are highly dispersed, and application scenarios are extremely diverse.
However, Rockchip’s semi-annual report released on August 19 provided numerical answers to how profitable this business can be. The financial report shows that the company achieved operating revenue of 2.046 billion yuan in the first half of the year, a year-on-year increase of 64%; net profit reached 531 million yuan, with a staggering year-on-year increase of 191%.
A company whose main business is not in the ‘mainstream’ track has seen rapid profit growth, which at least indicates that the real commercial value of the IoT market composed of automotive, industrial, and robotics may be underestimated by the outside world.
A Chip Serving Hundreds of Industries
According to Rockchip’s semi-annual report, the company achieved operating revenue of 2.046 billion yuan in the first half of 2025, a year-on-year increase of 63.85%; net profit attributable to shareholders was 531 million yuan, a year-on-year increase of 190.61%. In the second quarter alone, the company’s revenue reached 1.161 billion yuan, a quarter-on-quarter increase of 31.18%; net profit attributable to shareholders was 322 million yuan, a quarter-on-quarter increase of 54%.
The speed of profit growth significantly outpaces the speed of revenue growth, which usually indicates that the company’s products are being sold at higher prices, or that higher-margin products are selling more.
Data also confirms this: the company’s gross margin in the second quarter reached 43.3%, an increase of 6.5 percentage points year-on-year. Some brokerage analysts believe this is mainly due to the increased volume of mid-to-high-end products, which improved the company’s overall product structure.
In its financial report, Rockchip attributed its performance growth to ‘flagship products like RK3588 and new products like RK3576 leading the AIoT product lines to continue maintaining rapid growth, especially in key areas such as automotive electronics, industrial applications, machine vision, and various types of robotics.’
This raises a question: What exactly is the RK3588? How does it support Rockchip’s rapid growth in the first half of the year?
SoC, or system-on-chip, can be understood as condensing the motherboard of a computer, along with the central processing unit (CPU), graphics processing unit (GPU), neural processing unit (NPU), and various functions, into a single chip. It is the ‘brain’ of various smart devices. According to public information, Rockchip’s flagship chip RK3588 uses an 8nm process and an 8-core processor, and is equipped with a self-developed NPU with a computing power of 6 TOPS.
A specific example is the quadruped robot from Yushu Technology. At the 21st Shanghai International Automotive Industry Exhibition on April 23 this year, the Economic Observer reporter learned on-site that the core control chip of this robotic dog is Rockchip’s RK3588S chip.
According to the on-site staff’s introduction, the RK3588S chip plays the role of the central brain in the robotic dog, responsible for real-time motion control, multimodal perception processing, and AI inference core functions. Through efficient computing power allocation, this chip can reduce the response delay of motion control to less than 3 milliseconds. Additionally, this chip is compatible with ROS and ROS2, which are common development systems in the robotics field.
Yushu Technology is not a traditional consumer electronics manufacturer; it represents the rapidly growing emerging smart hardware market that Rockchip’s chips are targeting.
In the chip industry, these technical indicators do not represent everything. Rockchip’s choice is to combine these performances with an extremely rich set of interfaces, allowing this chip to adapt to hundreds of different industry applications. According to the information disclosed in the semi-annual report, the applications of RK3588 cover large-screen devices, edge computing, automotive intelligent cockpits, multi-camera devices, various robots, and financial payment devices.
This ‘one chip serving hundreds of industries’ model also determines Rockchip’s business model: unlike smartphone chip manufacturers that only need to serve a few large customers well, Rockchip needs to serve thousands of small and medium-sized customers distributed across different industries.
At the semi-annual performance communication meeting on August 22, Rockchip’s director, vice president, and CFO Wang Haimin stated in response to investor questions that the company adopts a sales model primarily based on distribution, supplemented by direct sales, to achieve wide market coverage by integrating external channel resources.
The foundation of this model is a large developer ecosystem. At the semi-annual performance communication meeting on August 22, Rockchip’s vice president Li Shiqin stated that from the early RK3188 chip to the current RK3588, each generation of products has been aggregating customers and partners. The specific approach is to provide a complete reference design platform and rich scenario-based solutions to lower the development threshold for engineers.
Li Shiqin summarized the resulting network effect as ‘the more it is used, the faster it iterates, and the wider it adapts.’
With products and an ecosystem in place, profit growth also requires a clear market direction.
When the Economic Observer reporter asked at the performance communication meeting which fields have higher demand certainty in the second half of the year, Rockchip’s board secretary Lin Yuqiu stated that automotive electronics, industrial applications, machine vision, and robotics are all key areas that have been laid out in recent years and have achieved good growth.
She particularly mentioned the robotics business, stating, ‘Robotics is a rapidly growing market this year and in the foreseeable future.’
It is worth noting that the semi-annual report shows that Rockchip also released a new generation of AI vision processor RV1126B in the first half of this year. According to Li Shiqin’s introduction at the performance communication meeting, this chip has an NPU with 3 TOPS of computing power and integrates an independent AI-ISP module.
ISP stands for image signal processor, which is mainly responsible for converting the raw data captured by the camera sensor into a clear image. The feature of Rockchip’s chip is that it integrates an independent AI-ISP module. According to the semi-annual report and Li Shiqin’s statement, this means that when running image tasks that require AI computation, this independent module can handle the processing without using the higher power-consuming NPU, thus achieving bandwidth savings.
This continuous product iteration and market expansion together form a complete picture of Rockchip’s performance growth in the first half of the year.
This choice of business model also determines its position in the industry. When people talk about the ‘king’ of a certain popular chip track, such as tablets, TVs, or security, Rockchip is usually not the first name that comes to mind. However, upon review, it can be found that almost every track has its profitable presence.
There Are Multiple Ways to Profit from AI
In the first half of 2025, the entire semiconductor industry is talking about growth.
According to the World Semiconductor Trade Statistics (WSTS) forecast, the global semiconductor market size will reach $700.9 billion this year, a year-on-year increase of 11.2%. Rockchip also cited data from the National Bureau of Statistics in its semi-annual report, stating that domestic integrated circuit production increased by 15.8% year-on-year in the first half of the year, and the revenue of the integrated circuit design industry increased by 18.8% year-on-year.
The engine of growth is almost pointing in the same direction—artificial intelligence.
The emergence of large AI models has raised new demands for computing power, and the industry’s focus is concentrated on products that can provide massive computing power, such as the most advanced GPUs and their accompanying HBM memory. This is a race with huge investments, a game for the giants.
Rockchip’s business landscape and company size determine that it is not currently the main player in this competition, but it also has its own way to profit from this AI feast.
Rockchip’s approach is the ‘edge computing co-processor RK182X.’
According to the company’s semi-annual report, the launch of this product aims to solve a specific problem: when customers need to add AI capabilities to existing products, they often face difficulties such as long development cycles and high costs. The role of the co-processor is to allow customers to retain their existing mature solutions. The semi-annual report mentions that customers can meet the AI computing power upgrade needs through ‘flexible pairing with the main processor,’ thus ‘greatly saving time costs and development cycles.’
At the semi-annual performance communication meeting on August 22, Li Shiqin introduced the specific performance and latest progress of this chip (RK182X). He stated that this chip can efficiently support mainstream parameters such as 3B and 7B for edge models and is currently in the ‘sample delivery stage.’
Behind this is a very realistic question: if one cannot produce the most advanced GPU, does that mean one cannot profit from AI?
Rockchip’s co-processor provides a negative answer.
The AI business is not solely about training large models. Once large models are trained in the cloud, they ultimately need to be deployed to thousands of terminal devices for specific applications. These devices, such as a smart lock or an industrial camera, do not require an expensive GPU capable of training models, but rather a chip that is cost-effective, low-power, and can run AI.
The co-processor targets this ‘AI application’ market. It does not aim to enter the ‘AI training’ track crowded with giants but chooses to provide a low-cost AI upgrade solution for thousands of existing customers in the IoT (AIoT) field where it has been deeply cultivated for many years.
There has always been a joke in the industry that Rockchip is the ‘Pinduoduo of the chip industry,’ doing business that large manufacturers look down on and small manufacturers cannot undertake.
This saying does not refer to product prices but rather a way of finding business: for example, when giants fiercely compete in the core business districts of first-tier cities, there are always some who choose to explore new territories in more segmented sinking markets.
Rockchip is such a company. While the industry’s attention is focused on smartphones, AI, and servers in these ‘core business districts,’ it conducts its business in those numerous but relatively silent fields.