Dark horses on the edge: AI concentration camps leading to the future

In the new wave of technology, edge intelligence is becoming more and more important

It represents a new computing paradigm that applies artificial intelligence or large-scale models to edge devices and sensors close to the data source, rather than relying on traditional cloud computing

Currently, edge artificial intelligence (AI) chips are mainly used in consumer electronic devices. Among them, high-performance mobile phones dominate the consumer-grade edge AI chip market. Edge AI is embedded into the main processor (AP) of mobile phones, but only a few giant companies (such as mobile phone manufacturers such as Apple, Samsung, and Huawei, and mobile phone AP suppliers such as Qualcomm, MediaTek, and Unisoc) have mastered the technology. The ability to integrate AI functions into APs makes it beyond the reach of most edge AI chip startups

However, more and more edge AI chips are being used in non-consumer devices and occasions, such as smart security, ADAS/autonomous driving, smart homes, wearable smart devices, as well as AI in public scenes, commercial and industrial scenes Applications, such as intelligent transportation, smart cities, factory machine vision, robots and AGV, etc. These emerging AIoT and industrial IoT application scenarios have brought more opportunities to many edge AI chip design companies, and venture capital investors have also realized the huge business opportunities contained in them. Therefore, whether it is the global or domestic market, more and more AI chip startups have successfully obtained financing

However, there are not many artificial intelligence chip companies that have achieved huge growth and successfully gone public in recent years, let alone gained recognition in the chip industry in mature markets such as the United States. Congchain Group (ICG), which successfully landed on the Nasdaq market in the United States this year, may be regarded as a rare and typical exception. It is gratifying that ICG has begun to set its sights on the field of artificial intelligence chips, which is an area with huge opportunities and is still in the early stages of development. At the same time, ICG is also aware of opportunities in the field of edge artificial intelligence that are both feasible and have long-term commercial application scenarios

According to data from STL Partners, the potential market for edge computing will grow at a compound annual growth rate of 48% over the next decade, from US$9 billion in 2020 to US$445 billion in 2030. Among them, edge infrastructure is growing the fastest. As we all know, edge AI chips are at the core of this field and have important strategic significance

According to data from Yiou Think Tank, China’s edge computing market size has reached 42.79 billion yuan in 2021. Among them, the edge hardware market size is 28.17 billion yuan, while the edge software and service market size is 14.62 billion yuan. It is expected that from 2021 to 2025, the scale of China's edge computing industry will grow at a compound annual growth rate of 46.81%. By 2025, the overall size of the edge computing market is expected to reach 198.768 billion yuan. In addition, because of major breakthroughs in artificial intelligence and large-scale models this year, almost all previous forecasts need to be re-estimated. In other words, under conservative assumptions, the scale of China's entire edge computing industry is expected to at least nearly double, and will soon enter the forecast range of 300-400 billion yuan. This mainly depends on the explosive growth of artificial intelligence applications in 2024

According to the famous "Andy-Beer Theorem", there is a spiral development and evolution relationship between intelligent terminal hardware and software. Generally speaking, software updates and upgrades should match hardware resources. However, the rapid development of generative artificial intelligence in recent years has led to huge changes and upgrades in artificial intelligence applications at the software and system levels. This also brings about an urgent need for "exponential" upgrades in the comprehensive performance of infrastructure and terminal hardware equipment. Because the Internet and the Internet of Things are interconnected, once key nodes undergo large-scale changes, other nodes will also change accordingly. This truth is both simple and profound

From the perspective of the industry chain, the core of edge AI is to introduce edge-side AI capabilities to further enhance the edge-side computing power and connection capabilities. The focus includes AI chips, computing power modules, edge gateways/servers/controllers and other hardware, AI algorithms/edge computing platforms and other software links.

In the core industry segment of the AI ​​field, AI chips refer to modules used to process a large number of computing tasks. These tasks are performed in artificial intelligence applications, while non-computing tasks are usually processed by the CPU. From the perspective of technical architecture, AI chips are mainly divided into three categories: GPU, FPGA and ASIC. Among them, GPU is a relatively mature general-purpose artificial intelligence chip, while FPGA and ASIC are semi-customized and fully customized chips designed for the characteristics of artificial intelligence needs. Generally speaking, typical AI calculations require CPU or ARM cores for scheduling processing, while a large number of parallel calculations are completed by GPUs, FPGAs or ASICs. For complex computing scenarios, the required computing power performance must be high, power consumption must be low, and cost-effectiveness must be considered in large-scale applications. At this time, choosing the ASIC chip architecture is the most appropriate choice. Therefore, in the context of the large model era, ASIC chips may be the technology that benefits the most in the field of AI chips

Major cloud companies around the world, such as Google, AWS, Meta and Microsoft, are committed to independent research and development of chips and adopt the design route of ASIC chips. At the same time, companies such as Qualcomm, Nvidia, ARM, AMD, Intel, MediaTek and Huawei are also constantly optimizing the functions of edge artificial intelligence chips. In addition, well-known foreign manufacturers such as NXP Semiconductors (NXP Semiconductors), Silicon Labs and ST (STMicroelectronics) have also begun to add edge artificial intelligence functions to MCUs or SoCs

In China, including Huawei HiSilicon, Baidu, Hanshiji, Jingjiawei, Horizon, Muxi Technology, Suiyuan Technology, Biren Technology, Alibaba Pingtouge, Rockchip, Amlogic Technology, Beijing Ingenics , Yuntian Lifei, Hengxuan Technology, Jiutian Ruixin, Hangzhou Guoxin, Espressif Technology, ICG (Conong Chain Group) and many other Chinese companies are laying out edge AI chip related fields

From an investment perspective, whether to choose a traditional well-known large factory or a start-up or growth-oriented enterprise, I believe that a basic feature of the future industry is that a hundred flowers bloom everywhere. Whether looking at landing scenarios or terminal equipment, there will be a trend of diversification, fragmentation, and more and more verticality. Therefore, the trend of chip customization will become increasingly obvious

From this perspective, the potential changes of those small and medium-sized enterprises that are flexible, elastic and focus on specific segments, or those enterprises that continue to expand across industries and categories with certain expertise advantages, may are larger, so they may have higher valuation expectations when the new economic cycle arrives

As for which application scenario will be the first to emerge in the future, it is difficult to predict in advance. The key is to identify those industry players who have already gained strength at the starting line. Their early deployment and layout must be of special significance. We must carefully study and dig carefully

In short, what this sentence means is that everything in the future is unknown, and we may all become dark horses. Through tracking and prediction, we can find clues to the dark horse, but we cannot be completely sure

The above is the detailed content of Dark horses on the edge: AI concentration camps leading to the future. For more information, please follow other related articles on the PHP Chinese website!