"The cloud platform in the intelligent era and the cloud platform in the mobile era before it will inevitably undergo significant changes. This issue, which has been the focus of the industry in 2023, is emerging after a year.
To study change, look at the people who are driving it. There is no doubt that in the field of cloud services, Baidu Intelligent Cloud is the most active force in China in the past year, taking the lead in doing a lot of exploration, such as on the basis of cloud computing resources, successively launched models, model development tools and application development tools. This activist, who is ready to show his strength in the intelligent era, recently announced the first annual report card of the Qianfan large model platform to bring to the AI application ecosystem.
On April 9, 2024, according to the information released at the first GENERATE Global Ecological Conference held by Baidu Intelligent Cloud, as of the end of March 2024, Baidu Intelligent Cloud's Qianfan large model platform is very active, providing services to 85,000 enterprises, helping them fine-tune 14,000 large models and develop more than 190,000 large model applications. In the past six months, the number of partners in Baidu Intelligent Cloud has surged fivefold, and more than 300 partner-developed apps have entered the market through the "Qianfan AI Native App Store".
It can be seen that the cloud platform ecology in the intelligent era is taking shape, and the Qianfan platform has gathered 120,000 startups and ecological partners, of which more than 55,000 are actively calling the platform API. In the first quarter of 2024, 8,100 startups and partners have developed AI-native applications using Qianfan AppBuilder, and more than 200 top customers have been served by large models.
Observing these AI application innovators, we can clearly see the transformation of the technology division of labor in the era of large models: the powerful generalization capabilities of large models and the emergence of MaaS (Model as a Service) have shortened the distance from technology to implementation, which has changed the relationship between cloud platforms and partners, and also put forward new requirements for traditional technology service providers.
In the context of the drastic changes in the industry caused by large models, various platforms and developers are rethinking their own positioning. For developers, they are worried about reinventing the wheel and being swallowed up by large platforms. In this case, Baidu, as a pioneer, provides an excellent observation window, and its observation and thinking have important reference significance for the majority of developers.
01 What should be the difference between the cloud in the intelligent era?
As early as September 5, 2023, at the Cloud Intelligence Summit, Baidu Intelligent Cloud took the lead in launching the first full-link large-scale model ecological support system in China. The system provides a range of support for 10 integrated partners, 100 application partners, and more than 10,000 start-up and agency partners, including AI-native application incubation, sales opportunities, marketing, and capability training.
The Ecosystem Strategy on April 9 continues the above four categories of ecosystem partners, but increases the market equity of partners (see chart above). The four types of partners can be further divided into two categories: one is called scenario co-creation partners, including comprehensive partners and application partners, and the other is user growth partners, including start-ups and agency partners. It is worth noting that in the past year, more than half of the delivery projects of Baidu Intelligent Cloud have been completed with "scene co-creation" partners.
A major strategic adjustment is the redivision of Baidu Intelligent Cloud into the large model market, subdividing it into the head market, the value market and the high-potential market. For customers in these three different markets, Baidu Intelligent Cloud and its partners have clarified their respective roles and cooperation methods, forming a more flexible, close, mutually beneficial and efficient cooperation ecosystem.
For customers in the top markets, the business scenarios they face are usually complex, with strong personalized requirements, and systematic solutions are required. Here, Baidu Intelligent Cloud hopes to find a partner with a strong customer relationship, a deep understanding of the customer's industry, and the ability to deeply integrate and deliver industry solutions and products, jointly tackle tough problems, complement each other, create typical application cases in the era of large models, and create new value for customers.
For the value market, there are more customers in this market than in the top market, and the business scenarios are more standardized and less complex. Partners can quickly replicate the industry benchmark created for leading customers, or provide relatively lightweight customized solutions. In this market, Baidu takes the partner first, does its own technical and business support, participates in the customized development of solutions on demand, and works with partners to develop markets and serve customers.
For customers in high-potential markets, their business scenarios are relatively simple, and they value plug-and-play solutions more. Here, partners will fully dominate sales and services, and Baidu will focus on polishing standardized products and solutions, and give enough concessions to ensure partners' profit margins and support partners.
The logic behind this change in the division of labor is that the ecosystem in the era of large models is different from the era of cloud computing, and the relationship between cloud platforms and partners is changing from traditional integrated services to deep co-creation partnerships.
In the traditional cloud era, AI application development is a "chimney". In the face of different scenarios, it is necessary to independently prepare data, train models, and develop applications, which have high thresholds and low efficiency, and in the face of blowout market demand, the traditional development paradigm is obviously no longer applicable.
Now, the powerful generalization capabilities of large models have emerged in the form of MaaS (Model as a Service), replacing the previous isolated "chimney" model, which greatly lowers the technical barrier. This opens up a wider market space for partners, and technical challenges are no longer a major obstacle. Now, it's more important to gain insight into customer needs and earn their trust.
Changes in technology will inevitably bring adjustments to the ecosystem to be built on the cloud platform. With the lowering of the R&D threshold and the improvement of efficiency, it can be said that the entire ecological environment has undergone fundamental changes, prompting the upstream and downstream of the ecosystem to re-divide labor and cooperate.
For example, with the technical support of the Qianfan model platform, Xinzhi Software has been able to provide 24/7 intimate AI assistants for hotel chains and retail brands, creating a new marketing model. To a certain extent, this reflects the full-stack technical capabilities of the Qianfan large model platform, and also gives full play to the ability of Newtouch Software to deeply understand the pain points of the industry and customer needs.
Image source: Baidu Intelligent Cloud
From the perspective of Baidu Intelligent Cloud, the mission of the cloud platform at this time is becoming more and more clear - it is clear that building a solid technical foundation is the key to a prosperous AI application ecosystem. For example, in the past year, Baidu Intelligent Cloud has focused on innovation at the computing power level, and has also spent a lot of energy to improve the model layer, application development layer, and the technical support of the application itself.
At today's ecological conference, Baidu Intelligent Cloud launched a new plan aimed at expanding the technology base and strengthening cooperation with partners, clarifying the role of Baidu Intelligent Cloud and partners in jointly building a technology stack. This is also an important aspect of the renewal of the ecological strategy.
- At the computing power level, partners can jointly build a computing power center with Baidu and share computing power operation services.
- At the model level, partners can use ModelBuilder to create industry-specific models, provide comprehensive solutions, and promote the development of vertical applications in the industry.
- At the application development level, partners can develop applications based on Baidu's technical framework or build their own middleware to achieve a wider range of application scenarios.
- For the developed app, Baidu Intelligent Cloud's AI-native app store will assist partners in product exposure and sales. Agency partners can also participate in resale to achieve benefit sharing.
At the same time, Baidu Intelligent Cloud has launched a series of recruitment and assistance policies to support and amplify this cooperation model to the greatest extent.
From the practice of Baidu Intelligent Cloud, it can be seen that the core capabilities of the cloud platform in the intelligent era are much more complex than the cloud in the previous era. In particular, it is necessary to continuously improve its general capabilities, and at the same time support in-depth co-creation with more partners in terms of mechanism and ecology, so that Baidu Intelligent Cloud can lower the threshold for the use of AI applications and accelerate the practical implementation and application popularization of AI technology. And this is the core value of the cloud platform in the intelligent era.
02 At the heart of the change is the new relationship between the platform and partners
At the end of March last year, Baidu Intelligent Cloud launched the Qianfan large model platform, which integrates the Wenxin large model and more than 70 domestic and foreign third-party large models for enterprise customers to use. With the continuous implementation of AI applications, in addition to providing cloud computing resources and large models, Qianfan Platform has also successively launched ModelBuilder and AppBuilder tools, further simplifying the model fine-tuning and personalized AI application construction work for developers.
The launch of the Qianfan large model platform is like providing a mirror for AI application practice, reflecting the application and effect of large models in the past year. According to the latest data from Baidu Intelligent Cloud, the activity of customers and partners in model invocation and application development is not always on the rise, and this fluctuation reflects the true face of the large model implementation process.
Line chart of the growth of the number of customers/partners|Image source: Baidu Intelligent Cloud
Chen Zhiruo, general manager of Baidu's intelligent cloud channel ecology department, explained that the initial growth stage after the release of Wenxin Yiyan reflects the market's keen interest in large models. However, after entering July and August, the market began to gradually return to rationality, "After three months of excitement and the signing of the strategic agreement, I may not have thought about what I should do with the large model, and I began to investigate and explore the various business lines internally, what does the large model do? 」
By the end of the year, with a clearer understanding of the application of large models, some partners began to launch integrated solutions or lightweight services in specific domains. For example, many industry partners have begun to focus on a track to make all-in-one machines, or to build some lightweight microservices or applications based on large models.
Earlier this year, marketing activity increased again as some partners began to promote proof-of-concept products to customers. Despite the challenges of pricing and market fit, this marks the continued evolution of products and services. The recent growth indicates that partners are beginning to think more deeply about pricing, service, and implementation, signaling that the market has entered a new phase of optimization and commercialization iterations.
Chen Zhiruo pointed out that partners in certain areas have made significant progress, especially in industries such as education, smart office, and e-commerce. They used scenario-specific data and knowledge to successfully implement large-scale model applications.
"In many cases, the external form of the product may not have changed, but the technology and user experience behind it have completely changed. For example, if you feel that the video you watch has not changed, but the production behind the video is a large model, and the birthday cake you ordered has not changed, but the owner of the cake shop is already using the large model to do demand forecasting and intelligent customer service. 」
For example, by changing the original business model, Xinhua Miaobi launched by Bot Intelligence not only serves the original B-end users, but also goes deep into the C-end scenarios, such as the writing of official documents of grassroots civil servants, etc., to realize the accumulation and realization of its original data.
These partners who have successfully eaten "crabs" in the era of large models are significantly different from those in the era of cloud computing. In the era of cloud computing, cloud vendors provide more computing power and other infrastructure, and rely on partners at the top layer to reach end customers, but in the era of large models, the technical capabilities of large models allow customers with data and scenarios to directly develop AI applications, reducing their dependence on traditional service providers.
For many partners who traditionally prefer to resell resources and develop simple services, if they want to remain competitive in the era of large models, they need to transform, on the one hand, to accumulate professional knowledge, and to do further fine-tuning, pre-training, and industry data annotation when calling general models, so that large models can truly solve industry problems, and on the other hand, they need to better understand customers and dig deep into the needs of specific vertical industry scenarios, so as to better help customers implement AI applications.
On the one hand, how can enterprises with scenarios and customers make good use of large model base technology for AI applications, and on the other hand, how can platforms or technology-based companies provide good technical services, which is the meaning behind Baidu's intelligent cloud ecological strategic transformation.
Image source: Baidu Intelligent Cloud
In the end, how to work with partners to deliver large model technology to thousands of industries and even thousands of households through AI applications is the ultimate mission of a cloud platform in the intelligent era.