In the period when AI implementation is fraught with intense competition, the industry needs new ecological builders.
By 2025, the progress of implementing large models in all industries has reached a fever pitch.
Recently released bidding data shows that in February 2025, the number of winning bids for large model - related projects was 101, with a total value of approximately 464 million yuan. Affected by the Spring Festival, this figure decreased compared to January. However, it is still higher than that of any month in the first half of 2024.
Almost at the same time, 36Kr learned that several To B Agent startups had revenues exceeding 10 million yuan in Q1 this year, nearly a five - fold increase compared to the same period last year.
However, there are not many suppliers capable of meeting these demands. "Customers have more and more demands, and there are also more customization requirements. Coupled with innovation demands and services, we can't handle many orders," admitted the aforementioned Agent company.
This is not a problem faced by just one company.
"After communicating with multiple partners recently, we believe that everyone is facing common challenges," said Wang Liping, the vice - president of Lenovo Group and the general manager of Lenovo's China Public and Business Group. Partners represented by ISVs are currently in urgent need of solutions to challenges such as technology and delivery to help customers implement AI in the last mile.
Wang Liping, the vice - president of Lenovo Group and the general manager of Lenovo's China Public and Business Group
Lenovo is trying to help these ISVs overcome the barriers in AI services and build a new AI ecosystem in various ways.
AI Needs New Ecosystem Builders
The current surge in demand for large model implementation started with DeepSeek.
During the Spring Festival, when this "AI catfish" from China, the world's largest, emerged as a dark horse and challenged the leading position of US AI, hundreds of enterprises around the world began to have more confidence in Chinese AI and were more willing to try AI applications.
Catalyzed by DeepSeek, various Agents such as Alibaba's Qwen and Manus have emerged explosively in the past few months, making 2025 a real year of AI.
"Now customers no longer need an explanation of what an agent is," said Gu Xuguang, the senior director of the Qingtian AI Technology Center of Lenovo's China Solution and Service Business Group.
Gu Xuguang, the senior director of the Qingtian AI Technology Center of Lenovo's China Solution and Service Business Group
The problem in the domestic market is that due to the cautious attempts of enterprises with large models in the past two years, many ISVs are temporarily unable to fully meet this wave of explosive demand.
In addition, technology solutions such as agents are still evolving, and the routes of reinforcement learning and edge - side models are also continuously changing. ISVs need to invest too much in learning and experimentation. These problems have become real obstacles for enterprises to implement large models.
"When launching a complete agent solution, everyone will face three major challenges," summarized Lu Yuan, the general manager of Lenovo's AI Workstation in the China Public and Business Group, after communicating with a large number of ISVs.
Lu Yuan, the general manager of Lenovo's AI Workstation in the China Public and Business Group
The first challenge is that the personnel implementing large models often need to have knowledge of both software and hardware, and testing resources need to be configured according to different GPUs. This kind of investment is beyond the reach of ordinary ISVs.
Second, the productization cycle of large models is long. For example, for products like all - in - one machines, which involve hardware selection, software - hardware adaptation, and customization, it takes a very long time to bring the solution to the market.
Third, the speed of converting solutions into orders is very slow. This requires precise sales channels, professional marketing planning, and rich promotion platforms.
Obviously, none of these problems can be easily solved by a small company. The AI industry needs new large - scale enterprises to take the lead in building an ecosystem.
Which type of large - scale enterprise is most suitable to be the co - builder?
From the perspective of the demand side, for enterprise customers to implement AI projects, not only must the products be of high quality and diverse, but also a trustworthy brand and timely and comprehensive services are essential.
On the product side, enterprise - level projects often involve private deployment, so service providers need to be able to smoothly adapt software and hardware. On the service side, enterprises require service providers to have a nationwide and timely response. On the brand side, well - known and experienced large - scale service providers are the first choice for customers.
Currently, among the large - scale enterprises capable of building an ecosystem in the AI field, cloud providers and some chip manufacturers are in the majority.
However, upon closer inspection, cloud providers have difficulty directly meeting the private deployment needs including all - in - one machines, and domestic chip manufacturers are still in the growth stage. They are not the ideal candidates.
Under such circumstances, Lenovo, a leading technology enterprise with both software and hardware capabilities, brand influence, and a marketing and service system, plans to build an AI ecosystem with its partners.
As a technology company established for 41 years, it has a thorough understanding of different types of hardware that can be adapted to various software, and its marketing and service networks cover the whole country.
At this time, it is most appropriate for Lenovo to take on the important task of building a new AI ecosystem.
Lenovo Empowers ISVs in Four Dimensions to Seize the AI Track
Lenovo's accumulated know - how in the software and hardware fields, meticulous service capabilities, and extensive marketing capabilities are the sources of its strength to take on this heavy responsibility.
For Lenovo, these are generally referred to as the "Four Forces" - "Brand Power", "Solution Power", "Marketing Power", and "Sales Power".
As an enterprise established for more than 40 years, Lenovo's brand influence is beyond doubt. In the past two years, it has also been continuously investing in the new - generation AI represented by large models, and has formed a rich product matrix of AI brands including servers, workbenches, and AI PCs, and has won wide - spread brand recognition.
Marketing power is also one of Lenovo's advantages. It can help partners quickly reach customers online and offline through differentiated channels.
"Nearly 2,000 partners participated in the Lenovo Conference offline in the past two days. Now we have nearly 13,000 offline partners across the country, and they have very extensive channels to reach customers," said Lu Yuan, the general manager of Lenovo's AI Workstation in the China Public and Business Group.
Solution power can be regarded as a highlight of Lenovo's efforts in the AI field. Relying on its in - depth knowledge of software and hardware, technology and product layout, and experience in serving customers, this leading company in the industry has refined products that are difficult to directly deliver into solutions.
Sales power also comes from Lenovo's years of accumulation. Lenovo is currently expanding customer bases jointly with its partners based on its sales network and also provides pre - sales support. Moreover, it can help partners with delivery.
Currently, Lenovo has approximately 4,400 maintenance stations and 24,000 certified engineers across the country.
"You may not be able to find a supermarket within ten kilometers in some places, but you can see a Lenovo maintenance station. As long as you use a PC or a Lenovo mobile phone, your service needs will be met," said Lu Yuan.
These "Four Forces" that can be empowered externally are exactly what ISVs need.
Now, for AI ISVs to serve a customer, they not only need brand support but also need to refine cutting - edge AI products into complete solutions and then promote them to the market.
"It takes at least half a year and at most one year for many partners to form a solution," Lu Yuan found that the speed of ISVs converting products into orders is very slow. To create new solutions and promote them to the market, they need diverse sales channels, professional marketing planning, and rich promotion platforms to reach target customers, but the development of these capabilities requires time and accumulation.
The difficulties faced by ISVs are exactly what Lenovo's "Four Forces" can make up for.
Qingtian AI Adaptation Center: One - Stop Support for ISVs Throughout the Lifecycle
Recently, Lenovo officially announced the Qingtian AI Adaptation Center, which concentrates the advantages and sincerity of the "Four Forces".
"The Qingtian AI Adaptation Center can quickly help partners deploy agent and large - model products on hardware to form all - in - one machines," Gu Xuguang, the senior director of the "Qingtian" AI Technology Center of Lenovo's Solution and Service Business Group, concisely explained its significance.
This is specifically tailored to the needs of Chinese enterprise customers. Considering efficiency, security, and procurement habits, the all - in - one machine model is the first choice for most Chinese enterprises.
The goal of the Qingtian AI Adaptation Center is to help ISVs quickly and reliably meet such demands. Specifically, the Qingtian AI Adaptation Center can bring three values to ISVs: software - hardware adaptation value, joint marketing value, and service sharing value.
First is the software - hardware adaptation value. It eliminates the biggest "roadblock" in front of many ISVs.
In practice, deploying large - model and agent all - in - one machines is not just a simple matter of installing software on hardware.
The simple phrase "software - hardware adaptation" hides countless pitfalls. For example, each GPU needs to be aligned with large models and large - model inference platforms. Once the versions are incompatible, there will be problems with the results. In addition, some GPUs may not support the quantized data format, which will increase the requirements for video memory, meaning more graphics cards are needed and the cost will be higher.
Moreover, smooth communication between GPUs and optimization of software - hardware collaboration require real "pit - stepping experience".
"We often see that theoretically, a certain GPU can achieve a throughput rate of 6,000 or 8,000 when running a certain model. But in reality, we find that its performance may be an order of magnitude lower. A lot of engineering optimization, such as operator fusion, must be done here," said Gu Xuguang.
Obviously, these details are cumbersome and complex for ISVs, and the input - output ratio of research is not high. The Qingtian AI Adaptation Center is Lenovo's quick - adaptation solution for this problem.
From a practical perspective, the Qingtian AI Adaptation Center divides the software - hardware adaptation steps into four steps: communicating requirements - solution adaptation - verification and testing - packaging and delivery.
In the requirement communication stage, Lenovo will communicate with ISVs about the scenarios and technical details required by customers. ISVs can configure all - in - one machines based on two dimensions: "cost - effectiveness of the computing power platform" or "specified GPU type".
"One way is that ISVs hope the overall solution has the best cost - effectiveness, the cheaper the better. Then we will recommend the hardware configuration with the best cost - effectiveness. It's like going to a restaurant, putting 200 yuan on the table and saying, 'Just make a combination for me as long as it tastes good'. The other way is to specify several GPUs, such as Muxi or Tianshuo. We can still make a suitable hardware configuration according to the requirements," Gu Xuguang introduced these two models by analogy.
In the solution adaptation stage, the Qingtian AI Adaptation Center can package and adapt the drivers, the model that performs best under the hardware support, and the inference platform according to the hardware configuration to achieve the best performance of the all - in - one machine.
"We have done a lot of such adaptations in advance. Every time a new large model comes out, we will retest the compatibility and performance through an automated testing platform. The same goes for GPUs. Many GPU manufacturers will send their cards to us for various compatibility tests. So the solutions recommended by Lenovo must be the optimal ones," Gu Xuguang revealed the secret behind the efficient adaptation solutions.
In the third step, verification and testing, the Qingtian AI Adaptation Center will conduct both functional and performance tests.
Functional testing means that Lenovo will use a professional testing team to test the inference ability, tool - calling ability, process integration ability, multi - round dialogue ability, etc. of agents based on various test sets. Moreover, Lenovo's test sets are customized "secrets" for enterprise scenarios, which can measure the matching degree of products in enterprise scenarios.
Performance testing refers to testing core indicators such as the throughput and concurrency of products. Lenovo will provide test reports for ISVs to help customers review.
The fourth step is production and packaging. After the testing steps are completed, Lenovo will package the solution into an image and hand it over to the factory for production.
"Lenovo will ensure the quality of the equipment, and more importantly, ensure long - term operation and maintenance services and long - term maintenance services for the equipment," said Gu Xuguang.
Software - hardware alignment can be regarded as Lenovo's "core skill". In addition, the Qingtian AI Adaptation Center can also conduct joint marketing and share services with ISVs.
Joint marketing refers to the agent marketplace in the Qingtian AI Adaptation Center solution.