In 2024, as many large models have been approved, the commercialization process related to large models will begin to be further accelerated.
Therefore, the intelligent hyperparameters are expected to comprehensively track and analyze the trend and progress of the commercialization of the large model on a monthly and quarterly basis, based on the data of the public large-scale model-related bid-winning projects, and provide valuable market insights for relevant enterprises and practitioners.
The "Monitoring Report on China's Large Model Winning Projects" will analyze the overall market size, industry coverage, project type, representative manufacturers, representative projects, geographical distribution and other dimensions, so as to depict the current situation of commercialization of the large model industry.
Q2 2024 Key Insights:
1. In the second quarter, the number of projects and the amount of disclosed bids increased significantly from the previous quarter. Compared with the first quarter of 2024, the number of projects in the second quarter increased by 267% quarter-on-quarter, and the disclosed winning bid amount increased by 162% quarter-on-quarter.
2. The number of application projects and the amount of disclosure have increased significantly. In the second quarter, the number of application-based large model bidding projects and the proportion of disclosed amounts increased significantly, indicating that the implementation of AI large models in specific scenarios has accelerated.
3. The implementation of AI large models in 6 industries is faster. In the second quarter, the number of large-scale model winning projects was calculated according to the number of projects, and the prominent landing industries include: education & scientific research, energy, operators, central state-owned enterprises, government affairs, finance, etc. During the quarter, these six major industries accounted for 80% of the number of disclosed projects.
4. Baidu, iFLYTEK, and Zhipu AI lead the industry. In the second quarter, Baidu, iFLYTEK, and Zhipu AI led the way in the number of large-scale model-related winning projects, showing industry leadership, but these three manufacturers also had challenges. With the efforts of traditional manufacturers such as Alibaba Cloud, Tencent Cloud, and Huawei, market competition will become more intense.
5. The landing area of the AI large model project has a wider coverage. In the second quarter, the winning projects were spread across 28 provinces and municipalities in China, covering a wider area, but the first-tier economic regions and major economic provinces are still the concentration of projects.
The following is the full text of the monitoring report for the second quarter of 2024:
1. The overall market
With the iterative upgrading of large AI models, the capabilities have reached a new level, and the commercialization of large models has entered the fast lane.
In the second quarter, the number of large-scale model-related winning projects counted by public channels was 187, of which 60 projects did not disclose the amount of winning bids, and the remaining 127 winning projects disclosed an amount of about 679 million yuan.
Compared with the first quarter of 2024, the number of projects in the second quarter increased by 267% quarter-on-quarter, and the disclosed amount of winning bids increased by 162% quarter-on-quarter.
As we said in our Q1 monitoring report, in the second quarter of 2023, although there were some large-scale model-related research projects in China's bidding market, the number and amount were negligible, so in the 2024Q2 report, we also omitted the year-on-year comparison.
In the first half of 2024, the disclosed amount of large-scale model-related winning projects has reached 939 million yuan, which has exceeded the whole year of 2023 (789 million yuan), exceeding by nearly 20%.
In the first quarter, we had expected a 150% increase in the size of the large model-related bidding market in 2024. According to the current trend of project bidding, this data is a bit conservative, and the growth rate of the overall market of more than 200% is very expected.
In the second quarter of 2024, the median amount of the disclosed amount of the large-scale model-related winning projects was 1.3 million yuan, while the average amount of the award-winning projects disclosing the specific amount was 5.3453 million yuan, which is related to the large amount of computing power projects related to some large models in the quarter.
From the perspective of intelligent hyperparameters, the median of 1.3 million yuan can better reflect the project status of the large-scale model bidding market at this stage - most enterprises or organizations are still in the stage of trying out a few scenarios, and the amount invested is relatively limited.
2. Large model project type
At present, there are many types of bidding projects related to large models, but on the whole, we divide them into four categories: computing power (computing resources required to run large models), data (large model-related data labeling or data resources, etc.), large models (various large model software and related support platforms), and application (the integration and implementation of large models in clear scenarios).
The categories of projects counted in the second quarter of 2024 are as follows:
From the above table, we can see that compared with the computing power projects in the first quarter (73.7% of the disclosed project amount), in the second quarter, the number of application large model bidding projects and the proportion of the disclosed amount have increased significantly.
This is actually very similar to the experience of practitioners, that is, after entering the second quarter of 2024, the implementation of AI large models in specific scenarios will progress faster. If the introduction of large models generates obvious value in the early business scenarios, this should further promote the rapid growth of the application bidding project market and form a mutually reinforcing "flywheel".
In the second quarter, the number of winning bids and the amount of disclosed bids in the data and large model categories were small, but these projects also increased overall. Subsequently, with the implementation of large models in vertical fields, data-related projects should grow significantly.
3. Industry distribution of large-scale model projects
In the second quarter, the number of winning projects in the large model was calculated according to the number of projects, and the prominent industries include: education & scientific research, energy, operators, central state-owned enterprises, government affairs, finance, etc.
Similarly, the number of disclosed projects in the above six major industries also accounted for 80%, and government and energy customers were the largest buyers of large models, accounting for more than 50% of the total amount of disclosed projects in the two industries.
Although large models have also been implemented in transportation, tobacco, media and other fields, in the second quarter, AI large models have been implemented faster in six major industries, including education, energy, operators, central state-owned enterprises, government affairs, and finance.
Industries in these industries are usually rich in data, and also have the resources to make tentative investments in advanced technologies such as AI large models, or conduct research and exploration.
4. Winning bidders and representative projects
In the second quarter of 2024, among the 187 projects counted, 144 won bids.
Winning bidders in the second quarter include: carriers, universities, research institutions, cloud vendors, AI model start-ups, and small and medium-sized integrators. On the whole, in the second quarter, regional integrators became the main bidders in the regional market.
In terms of manufacturers, in the second quarter, Baidu, iFLYTEK, and Zhipu AI were the high-frequency faces among the winning bidders, showing the leading position in the industry. However, it can also be seen that Alibaba Cloud, Tencent Cloud, Huawei, Volcano Engine, SenseTime and other manufacturers are also making efforts.
In the second quarter, Baidu continued its strong growth momentum since the beginning of the year, which is related to its first-mover advantage in the field of large models. However, it is an undeniable fact that the first-mover advantage of Baidu's large model technology is shrinking, and it may be necessary to use the landing value of more projects obtained in the early stage to reflect differentiation.
In the second quarter, iFLYTEK showed strong competitiveness in central state-owned enterprises, operators, financial and other industries, and won large orders of more than 10 million yuan from customers such as Guoyuan Securities and China Mobile Sales Branch. However, it is worth pointing out that iFLYTEK does not belong to a large factory with a strong background, and the continuous large-scale model competition will lead to its R&D expenses continuing to be high, and it is difficult to contribute profits to the large-scale model winning project, which will affect the performance of iFLYTEK as a listed company.
In the second quarter, the number of winning bids of Zhipu AI also achieved industry-leading, but in comparison, the industry distribution of its projects is relatively scattered, and the 10 projects counted are distributed in transportation, energy, operators, central state-owned enterprises, government affairs, finance, media and other fields, which seems to be very unfocused, and geographically scattered. Even though the field of large models is in the stage of building industry benchmark projects, this way of expanding the market is certainly not low for a start-up.
5. Regional distribution
In the second quarter, the winning projects were located in 28 provinces and municipalities directly under the central government in China, mainly in Beijing, Sichuan, Shanghai, Guangdong + Shenzhen, Henan, Zhejiang and other places.
Compared with the first quarter, the second quarter covered a wider region, proving that the AI model has a wider geographical coverage. However, on the whole, projects are still concentrated in first-tier economic regions such as Beijing, Shanghai and Guangdong, and projects in other major economic provinces are also growing rapidly.
Note: In the second quarter, there are 60 projects that have not disclosed the amount of winning bids, in order to facilitate statistics, the project amount is calculated as 0, in addition, the project classification, industry classification, etc. are not strict, which may affect the final statistical results, please be aware of the readers.