news

Big model technology has swept the global technology community, and China has quickly followed this trend and started to implement it in many fields, improving corporate operational efficiency and consumer experience. However, big model companies also face challenges such as technology maturity, cost, data security, industry differences, and user acceptance. Commercialization is still a problem facing all manufacturers.

After more than a year of practical application, the big model has begun to take shape in the bidding market, and we have also seen many interesting winning projects, even though the To C commercialization prospects are still unclear.

For example, the large model base and application of rice technology projects, the regional large model of weather forecast, the pre-diagnosis large model, and video generation R&D projects, etc., the large model technology is gradually "penetrating" into more industries.

According to the data provided by Zhiliao Biaoxun to Silicon Star, in 2023, the bidding market launched 190 large-scale model procurement demands, with a procurement scale of 595 million yuan. A total of 135 buyers and 131 suppliers participated in the transaction throughout the year, of which 25.92% of buyers had annual procurement demands of more than 2 times.

In the first half of 2024, the bidding market has generated 498 large-model related bidding projects with a bidding amount exceeding 1.34 billion yuan, which has exceeded the full-year scale of 2023.

Among them, the project with the highest amount is the Guizhou East-West Computing Large Model Construction Project, with a bid amount of 176.7364 million yuan. The winning company is Yunshang Kunpeng, whose controlling shareholder Tuowei Information is a company in Huawei's industrial chain.

In terms of market geographical distribution, in the first half of 2024, procurement demand is mainly concentrated in Beijing, Guangdong, Shanghai, Jiangsu, Zhejiang and other places. Guizhou Province has become the province with the highest procurement amount thanks to the projects related to the construction of the East-West Big Data Model.

In terms of the proportion of purchase times by various types of buyers, private enterprises accounted for 24.7% and the amount of purchases made by private enterprises accounted for 45.49%.

In terms of the number of winning bids by suppliers, Zhipu AI, Baidu, iFLYTEK and Huawei ranked at the top.

Silicon Star also combined public channels such as Zhiliaobiaoxun, China Government Procurement Network, and Xunbiaobao to sort out some of the winning bids for large model-related projects this year.

According to incomplete statistics, Zhipu has won 18 projects this year, covering multiple fields such as finance, education, energy, communications, and medical care. The purchasers include state-owned enterprises such as China Mobile and China Telecom, financial institutions such as China Merchants Bank, as well as educational institutions and research institutes. The project investment scale ranges from 80,000 yuan to 4.99 million yuan.

Among them, Migu's video generation R&D project usesTransformerThe model performs video diffusion and multimodal video generation, demonstrating Zhipu’s progress in cutting-edge video generation technology.

According to incomplete statistics, Baidu has won 17 projects in the big model, covering multiple fields such as medical care, finance, energy, environmental protection and transportation. For projects with higher bid amounts such as the "Trade Union Intelligent Construction Project" (31.58 million yuan) and the "Intelligent Middle Platform Intelligent Center Research and Development" (4.767 million yuan), Baidu has cooperated with many large state-owned enterprises and industry leaders, such as China Tower, China Southern Power Grid, China Merchants Group and China Unicom. According to incomplete statistics, in the field of intelligent Internet of Vehicles, Baidu is the only company that has won the bid for a transportation big model-related project.

According to incomplete statistics, iFlytek has won bids for 15 large-model related projects this year, covering multiple fields such as finance, education, energy, communications, and justice. The winning projects include both infrastructure construction, such as the "AI Basic Large Model and Data Training Platform" of the State Energy Group, and specific application development, such as the "Digital Assistant Innovation Project Based on Large Models" of the Beijing Open University. In addition, many projects also include iFlytek's hardware products.

Among them, the most eye-catching is that Sanya Yazhou Bay Science and Technology City Development and Construction Co., Ltd. invested 27.0496 million yuan in the purchase of cognitive large model base facilities and innovative applications for the "National Salt-Alkali Tolerant Rice Technology Innovation Cooperation Platform Project". Not only is the amount large, it also demonstrates the importance of AI technology in the field of agricultural science and technology.

The project content includes "1 general cognitive big model + 2 general big model support platforms (big model training platform, big model application development platform) + 1 data training task + 5 types of big model application scenarios (agricultural research assistant, department digital employee, Zhiwen Encyclopedia, work order service consulting assistant, public service matters consulting)".

The projects Huawei won bids for covered industries such as energy, finance, and meteorology. Among them, the Shenzhen Weather Forecast Regional Model Project, a collaboration between Huawei and the Shenzhen Meteorological Bureau, won the bid with a bid of 2.8 million yuan. Last July, Huawei Cloud Pangu Meteorological Model was featured in Nature. Huawei's big model strategy has always been oriented towards the ToB and ToG markets, emphasizing in-depth industry and solving industry problems.

The projects won by SenseTime cover important fields such as electricity and finance, demonstrating the company's technical strength in these vertical industries. The project amounts range from millions to tens of millions, reflecting the needs of AI applications of different scales. It is worth noting that SenseTime not only provides customized industry solutions, but also engages in infrastructure services, such as providing GPU resources for China Merchants Bank.

Alibaba Cloud, Tencent Cloud and Volcano Engine are in a similar situation. The number of related winning projects in public information is relatively small, which is closely related to their positioning as infrastructure providers. Cloud service providers have technical strength and resource advantages, and mainly focus on providing underlying computing resources, data storage and network services, rather than directly participating in the delivery of specific large model projects.

Unlike Zhipu, which is active in the large-scale model procurement market, the other three of the "Four Little Dragons" of large models have only a few involvements in this field.

MiniMax did not appear in the public bidding information this year. Last October, Shanghai Xiyu Technology (MiniMax) won the bid for China Merchants Bank's pre-trained basic large language model (millions of billions of dollars) procurement project. Other candidates included Baidu and Zhipu.

Persist in being a To C super appDark Side of the MoonIt did not appear in the procurement list, but Yang Zhilin's first entrepreneurial company, Circular Intelligence, won two procurement projects of Bank of Beijing. Last year, it won the bid for the "Bank of Beijing Intelligent Assisted Insight System AI Model Management" and other projects. Circular Intelligence CEO Chen Qicong once publicly stated in a speech that Dark Side of the Moon mainly undertakes toC and toD tasks, while Circular mainly undertakes toB tasks.

Circular Intelligence started commercialization in 2018, and its winning projects are concentrated in leading enterprises in the financial industry such as banking, insurance, and securities. For example, Circular Intelligence has cooperated with Bank of Beijing on at least 9 projects, and also posted a tweet on its official website that it had won the bid for Xiamen International Bank.

According to public information, Circular Intelligence and Dark Side of the Moon currently have a deep strategic partnership. Based on Dark Side of the Moon's universal big model, Circular Intelligence will provide the industry with better industry big model solutions and applications for various business scenarios. In addition, Circular Intelligence also emphasizes the long text capabilities of big models.

The other two companies among the big model star startups, Zero One Wanwu and Baichuan Intelligence, did not appear in the public bidding information.

Large model startups have made few moves in this field, which may be related to the complex and diverse demands of the domestic enterprise market, the fragmented industry market, and the large number of small and medium-sized customers. These customers have a strong demand for customized solutions, making it difficult to quickly standardize and modularize AI products. In the ToG market, although competition for intelligent projects is fierce, profits are meager, and government and enterprise organizations prefer to introduce packaged AI, cloud, IoT and other software and hardware solutions. Therefore, more companies choose To C as a breakthrough point.

The rise of big models has brought new business models and technical challenges. From market competition to project implementation, whether it is ToC, ToB or ToG, major manufacturers are moving forward in constant exploration.

At a time when the big model business model is still unclear, commercial realization through bidding has become an option. The advantage of this method is that it can quickly obtain projects and income, reduce market uncertainty, and enhance brand awareness and trust through cooperation with the government and large enterprises. However, the bidding process is cumbersome and competitive, with limited profit margins, and the project requirements are diversified and highly customized, which may lead to the dispersion of resources and make it difficult to form economies of scale and standardized products.

The AI ​​wave started around 2015. The four AI dragons that grew out of computer vision technology mainly made profits by providing customized AI solutions to enterprises and governments. Corporate and government bidding and centralized procurement became an important source of income for them, but also an important reason for their lack of follow-up.

In the foreseeable future, big model technology will play a role in more areas. In this industry where technology, market, and business model are all changing rapidly, its future direction remains to be seen.