Zhitong Financial APP learned that Guojin Securities released a research report sayingAI has landed deeply in many vertical scenes, such as finance, medical care, transportation, security and so on, and the expansion potential of application scenarios is strong, and the associated data is growing massively.According to the Institute of Information and Communication, since OpenAI launched GPT-3 in 2020, the number of parameters of super-large pre-training model and the scale of training data have increased according to the trend of 300x / year, objectively promoting the exponential expansion of the demand for high-quality data. The AI model gives rise to high requirements and new demand, and professional data sets and AI trainers demand high-quality professional data solution providers such as 688787.SH.
The main points of ▍ Guojin Securities are as follows:
AI basic data service providers are born out of the needs of professional data acquisition and division of labor.
The domestic AI basic data service industry mainly includes three industrial links: upstream-data generation and capacity resources, mid-stream-training data production, and downstream-AI algorithm research and development. Data engineering is the basic link of AI engineering. At present, it is at the key point of the market pattern becoming clearer, the iteration of new and old technologies, and the accelerated release of downstream demand.
The amount of AI rapid landing stack data has jumped exponentially, and the domestic scale of 10 billion is expected in 2025.
On the scene side, AI has landed deeply in many vertical scenes, such as finance, medical treatment, transportation, security, and so on, and the expansion potential of the application scene is strong, and the associated data is growing massively.
On the product side, it is believed that since OpenAI launched GPT-3 in 2020, the number of parameters of super-large pre-training model and the scale of training data have increased according to the trend of 300x / year, objectively promoting the exponential expansion of high-quality data demand.
According to iResearch, China's AI basic data service market may reach 10.1 billion yuan in 2025, double that in 2022.
The AI model gives rise to high requirements and new demand, and professional data sets and AI trainers need to benefit high-quality professional data solution providers such as Haitian AAC.
In the era of AI big model, unsupervised / semi-supervised training has returned to the mainstream, "basic model + fine-tuning" has become a new paradigm of AI development, and RLHF fine-tuning technology has given rise to two new requirements.
First, the pre-training stage gives birth to the demand of AI manufacturers for all kinds of professional scene data sets that are well marked and cleaned; the other is the interactive reward stage gives rise to the demand of AI manufacturers for artificial intelligence trainers with professional fact judgment and normative value judgment.
At present, the two types of emerging demand have not been fully and effectively supplied by the market. With the super-large-scale pre-training model promoting the exponential jump of training data, the blue ocean of the market urgently needs to be filled.
The demand for autopilot training data is in the ascendant, Haitian AAC power layout.
Autopilot accounts for about 35% of China's AI basic data service market, which is the largest downstream scenario, and the market size may reach 2.5 billion yuan in 2025. The business types in the field of self-driving can be divided into 2D image data annotation collected by vehicle camera and 3D point cloud data annotation collected by lidar. At present, customers in the industry have a high demand for full-stack closed-loop data solutions.
The concentration degree of the domestic market tends to converge, and Haitian AAC has a leading advantage.
With the demand-side AI vertical scenario and professional demand highlighted, the data demand increases, but the threshold of data acquisition business is raised; the supply-side domestic data security standards and related qualification requirements are improved, and the relevant solution capabilities are gathered to the head manufacturers.
Brand data service providers represented by Haitian AAC and Appen will replace small and medium-sized suppliers as the main supply force in the market in the future. According to IDC, Haitian AAC ranked first in the domestic AI basic data service industry with a market share of 12.9% in 2021. Appen has more advantages in coverage area, language / dialect coverage and so on.
Overseas benchmarking: Appen, the global leader in AI basic data services.
In the past five years, the company's revenue CAGR has reached 27.2%. The main business includes three categories: data collection, data preprocessing and model evaluation. The main customers include Google, Amazon, Microsoft and other well-known IT giants, customer quality is high. From a global point of view, Appen ranks first in the global AI basic data service industry in terms of income, staff capacity and labeling technology.
Stones from other mountains can be used to attack jade. for domestic manufacturers, actively clamping global AI head customers will help to obtain business dividends for large model pre-training; continuously expanding business matrix and opening up a model evaluation module to actively mine customer data requirements will help to deepen customer value vertically. Tamping data acquisition technology, following the industry trend, advanced data acquisition technology, including machine marking, will help to build their own business moat.