OpenAI has always been a bellwether in the AI industry.
The emergence of ChatGPT has enabled AI to usher in the "iPhone moment" it has been looking forward to; With the advent of Sora, large models began to evolve from unimodal to multimodal.
Now, the company has launched GPT-4o, which can accept any combination of text, audio, and image inputs and outputs, enabling more natural human-computer interactions.
Behind the increasing sophistication of large models is the exponential surge of parameters, the infinite "hunger" for computing power, and the huge power consumption, so the cost remains high.
However, in many cases, we only need a small amount of real-time intelligent analysis, and we do not need large models to process massive cloud data, which creates cost pressure in vain.
At this time, the advantages of small models are highlighted, such as the Phi-3 mini trained by Microsoft, which can be used in a computationally constrained inference environment, which is more convenient, more economical, and has low latency.
The computing power required for large models is mainly provided by AI chips such as GPUs and CPUs produced by NVIDIA, AMD, Huawei Ascend, Haiguang Information, Cambrian, Loongson Zhongke and other companies.
Small models are expected to be equipped with edge AI chips to make up for the limitations of the applicability of large models in different scenarios.
So, what is an edge AI chip?
Edge is a concept corresponding to the center, which refers to big data and cloud computing, which simply means that all data and computing resources are centralized, and when we issue an instruction on the PC, these resources will be mobilized.
However, we are now emphasizing the Internet of Things (IoT), which means that terminals are no longer just PCs and servers, but also smartphones, wearable devices, environmental monitoring devices, VR, and other products that can be networked, resulting in some scenarios of small data and real-time processing.
With the increase in the type and penetration rate of terminal equipment, edge AI is expected to usher in a golden period of development, and the compound annual growth rate of the market size of the edge computing industry in mainland China from 2021 to 2025 is expected to be as high as 46.81%.
Qualcomm launched the Snapdragon 8 Gen3 and Snapdragon X Elite with edge-end AI capabilities at the Snapdragon Summit 2023.
The domestic companies with more leading edge AI chip layout include Rockchip, Hengxuan Technology, Espressif Technology, Xinhai Technology, Allwinner Technology, Jingchen Shares, etc., which are expected to usher in a Davis double click in valuation and performance.
Among them, Rockchip's product performance can be said to be in the first echelon in the world, even compared with Huawei HiSilicon, Baidu, Alibaba, BYD, etc. are all company customers.
Rockchip's main product is intelligent application processor chips, which are edge AI chips, accounting for 89.52% of revenue in 2023, including a series of models such as RK3588, RK3399, RK3288, RK3528, RK3588 is the latest generation.
According to the different markets of downstream applications, the company's products can be divided into two categories: consumer electronics and intelligent IoT. Among them, consumer electronics are mainly for smartphones and computers; The scope of intelligent IoT is large, including drones, automotive electronics, intelligent robots, etc.
The advantages of Rockchip are mainly reflected in two aspects:
- Product performance advantages.
We compare Rockchip's latest generation RK3588 chip, and we can find that it is basically ahead of Jingchen Co., Ltd., Allwinner Technology and MediaTek in terms of chip computing power, audio codec, video codec, and highest display output.
Only in terms of computing power and maximum display output is slightly inferior to Huawei HiSilicon, Rockchip Micro is 6Tops and 8K60FPS, respectively, while HiSilicon is 9Tops and 8K120FPS.
In addition, the compatibility of Rockchip chips is also quite excellent, and it can support Android, Linux, RTOS, AAOS and a variety of domestic operating systems, and even various extended versions of Linux. Good compatibility means that the company's products are easier to promote.
From 2019 to 2023, Rockchip's R&D expense rate will exceed 20%, reaching a maximum of 26.36%, exceeding 500 million.
You must know that in the semiconductor industry, the R&D expense rate of more than 25% is actually not much, GigaDevice, Puran shares, Zhuosheng Micro, Weir shares, and even North Huachuang and SMIC, all of which are less than 20%.
- Customer Benefits.
With good products, customers will naturally come in a steady stream.
So far, Rockchip has thousands of customers, including Alibaba, Baidu, Tencent, Lenovo, Sony, Xiaomi, Midea Group, BYD, OPPO and other well-known companies at home and abroad.
For chip companies, customers are the guarantee of sales, even if the consumer electronics market enters the destocking stage from 2022 to 2023, which greatly affects the company's performance, but the revenue can be maintained at more than 2 billion.
And in the first quarter of 2024, there will be an inflection point, achieving revenue of 543 million yuan, a year-on-year increase of 64.89%, and a net profit of 68 million yuan, a year-on-year increase of 468.09%.
Don't look at Rockchip's net profit in the first quarter is less than 100 million, the company's profitability is not weak, and the gross profit margin from 2019 to 2021 will remain at 40%. It's just that after 2022, the downstream demand will be weak, and the gross profit margin will decline, but there has been a significant upward trend.
Rockchip's products are used in consumer electronics and intelligent IoT, and the future growth also largely depends on the development of these two markets.
First, consumer electronics is recovering.
In previous articles, we have repeatedly mentioned that the destocking of the consumer electronics market is nearing the end, and domestic smartphones in the first two months of 2024 increased by 14.6% year-on-year.
Coupled with the blessing of AI, the replacement boom caused by AI mobile phones and AI PCs, consumer electronics is expected to enter a new round of upward cycle.
Second, the IoT market is growing rapidly.
In recent years, IoT devices such as VR, MR, smart home, wearable devices, automotive smart cockpits, and intelligent driving have developed rapidly.
It is estimated that the number of IoT devices worldwide will grow from 14.4 billion to 29.7 billion in 2027, with a compound annual growth rate of 16% from 2022 to 2027.
The growth of these IoT devices, driven by intelligence, will fully drive the demand for the company's chip products.
Overall, edge AI chips are a market with huge potential, and Rockchip is one of the leaders, and its performance has also ushered in an inflection point, thus attracting 287 institutional positions (at the end of 2023), and the National Fund is still the company's major shareholder.
The above is only for the analysis of listed companies and does not constitute specific investment advice.
Follow Feijing Investment Research (id: Feijingtouyan) and receive the most valuable "Growth 20: 20 Companies with the Most Growth Potential in 2024" investment report now!
Source: Flying Whale Investment Research