【Abstract】In the current wave of artificial intelligence sweeping the globe, the battlefield of computing power is extending from the cloud to the edge.
The urgent need for massive data, real-time response, privacy security, and cost-effectiveness has jointly driven the rise of edge AI, where intelligent terminals require more “local intelligence”.
In this process, being the first to build a complete edge AI chip matrix is a prerequisite for occupying a more important position in the global AI landscape.
Search and add the WeChat account Aichengliang Andrew_7251 for in-depth discussions on more semiconductor industry developments. For market and project consulting, talent services, and decision-making analysis, add Aristodemus0403.
The following is the main text:
01
The Rise of Edge AI
From the mechanical revolution, electrical revolution, to the information revolution, and now the intelligent revolution, whether a technological revolution can truly change the development process of human society is tested in people’s daily lives.
Currently, whether it is the pioneer ChatGPT or the rising star DeepSeek, amidst the emergence of large models, the explosion of AI applications is still concentrated in the data center field, with the cloud becoming the focus of competition among various manufacturers.
However, for AI to achieve large-scale implementation, relying solely on the cloud is not simple.
For example, in the field of autonomous driving, the data collected by onboard cameras, millimeter-wave radars, and LiDAR is substantial, making it difficult for cloud data centers to meet the real-time response requirements of autonomous driving through information judgment and command transmission.
As the market demands higher real-time, privacy, and economic requirements for AI applications, the combination of “cloud-edge-terminal” becomes the key to the leap in AI application capabilities, based on which the status of edge computing is gradually rising.
In terms of intelligent terminals, due to cost and power consumption limitations, light intelligence has become a common choice for manufacturers. However, against the backdrop of continuous product performance improvement and the maturity of large models, the upgrade and optimization of devices face high costs and operational difficulties.
At this time, adding edge nodes can not only enable intelligent upgrades of terminals in small areas but also run large models under the design of light intelligence, which is a significant benefit for downstream fields such as terminal devices.
With over a decade of industry exploration gradually maturing, the emergence of large models has made algorithm lightweight possible. As the core hardware, chips are continuously enhancing their computing power and expanding bandwidth, further supporting the operational efficiency of large models and promoting the development of edge technology.
In this context, the domestic company Aisino has officially established edge computing as an independent product line, becoming one of the first to join the ranks of “shovel sellers”.
02
The Three-Stage Rocket Model
As a chip company, Aisino focuses its AI layout on the edge and terminal sides.
The company’s founder, Dr. Qiu Xiaoxin, has stated that edge and terminal chips, as upstream industries, can leverage the power of the ecosystem to achieve nonlinear amplification of utility, unlocking a product ecosystem worth hundreds of billions of dollars and thousands of applications, with a strong leverage effect.
Compared to the relationship between cloud-side chips and data centers, the application scenarios of edge computing are diverse, which requires manufacturers to provide downstream customers with more efficient, cost-controllable, and power-controllable computing platforms.
On this basis, Aisino’s internal strategy follows the “three-stage rocket” model of technology-product-application, defining market-demand-adapted products with leading technology and achieving closed-loop value through commercial sales.
From a technical perspective, similar to the fields requiring intelligent visual processing chips such as cameras, robotic vacuum cleaners, and smart cars, the development of AI edge computing also relies on intelligent perception processing and intelligent computing.
Since its establishment in 2019, the company’s core technology has undergone multiple iterations, with its self-developed core technologies—Aisino Smart Eye AI-ISP and Aisino Tongyuan mixed-precision NPU occupying a leading position in the fields of image and computing, providing foundational support for AI application innovation.
From the product definition perspective, the company has achieved a commercial closed loop from 0 to 1 in three major fields: intelligent terminals, edge computing, and assisted driving over the past six years, and is now focusing on the fourth growth curve in robotics, especially in the field of embodied intelligence.
Dr. Qiu Xiaoxin stated that currently, the domestic robotics field is still in the early stages of commercialization, with related companies still in the stages of robot body research and large model design, and large-scale commercialization will take another three to five years.
However, once the safety and intelligence levels of robots are improved and they successfully enter the home market, their commercial scale may even surpass that of the automotive market. Therefore, the company is also making early layouts in the robotics field in terms of technology and know-how.
For the AI industry, it is difficult for any single company to achieve universal intelligence. Therefore, Aisino, as a Tier 2 player, hopes to establish a “Swiss Army Knife”-style AI ecosystem community, creating diverse downstream applications with partners based on its core technologies and chip products.
It is reported that the company has established an open-source AXERA-TECH community and developed multiple developer kits with partners. To achieve the ease of use comparable to NVIDIA’s CUDA, Aisino will migrate and deploy open-source models to its chips, allowing users to train models using their own data to create vertical algorithms at low cost and high efficiency.
03
Opportunities for Domestic Products
Today, in the field of edge and terminal AI, domestic manufacturers are actually on the same starting line as overseas companies.
China and the United States are the two major highlands in AI development, but there are certain differences in their development directions.
While American semiconductor companies focus on high-end chips, data centers, and high-speed interconnects in high computing power fields, the domestic market has more advantages in equipment manufacturing, and customers are more willing to experiment with emerging technologies and products, making the application scenarios for AI chips richer and more feasible.
For example, the rapid progress of China’s intelligent driving chips is attributed to the development of domestic electric vehicles, forming a closed loop between scenario applications and technology products.
The closer AI chip companies are to application scenarios, the more accurately they can grasp customer needs, and the faster they can achieve leading advantages in rapid product iterations. From this perspective, the Chinese edge and terminal AI chip market is likely to give birth to world-class chip companies.
In addition to Aisino, companies like Rockchip and Allwinner Technology are also actively laying out in the field of edge AI chips.
In terms of development progress, Aisino’s edge computing product line will begin to be deployed in 2024, with the company’s shipment volume in 2024 already ranking among the industry leaders, and revenue in this field maintaining a high growth rate.
Currently, the company’s edge chips are mainly applied in various video-based intelligent applications and have established ecological cooperation with several leading software and hardware companies in the industry, achieving annual shipments in the millions.
In the field of smart cars, Aisino’s edge AI chips assist intelligent cockpits in running large models, creating AI Box co-processors that connect to intelligent cockpit domain controllers via PCIe or network ports, enhancing the AI computing power of the cockpit, supporting human-machine dialogue, and possessing multi-modal perception and reasoning capabilities, enabling on-demand upgrades of intelligent cockpits.
In the future, Aisino is expected to build a complete edge AI chip product matrix in the process of developing its independent product line, becoming a strategic supplier in the industry.
04
Conclusion
As a key node connecting the physical and digital worlds, the strategic value of edge computing is self-evident.
Whether it is intelligent visual processing chips or the development of independent edge AI product lines, Chinese semiconductor companies need more precise definitions at this critical juncture of the intelligent revolution.
Relying on its self-developed AI-ISP and mixed-precision NPU core technologies, as well as the technology-product-application three-stage rocket driving model, Aisino is accelerating the construction of a complete map covering intelligent terminals, edge computing, and future robotics.
As Dr. Qiu Xiaoxin has insightfully pointed out, the edge and terminal AI market in the Chinese market, which is closer to application scenarios, will also become fertile ground for nurturing world-class chip companies.
In this process, being the first to build a complete edge AI chip matrix is a prerequisite for occupying a more important position in the global AI landscape.
Search and add the WeChat account Aichengliang Andrew_7251 for in-depth discussions on more semiconductor industry developments. For market and project consulting, talent services, and decision-making analysis, add Aristodemus0403.
– XINLIU –
If you like it, please give Aisino a “👍” and “look”~