Text/Huang Haifeng's correspondence life

4月9日，阿里云AI势能大会在北京召开。在本次大会上，北京市政府副秘书长杨烁、阿里云智能集团资深副总裁刘伟光、地平线创始人&CEO余凯、阿里云智能集团产品生态发展部副总裁张亮等发表重要讲话，为AI发展趋势与实践应用提供建议。

"The social value of large models is being released in the enterprise market, and Alibaba Cloud will firmly invest in building full-stack leading technologies, continue to open source, and accelerate AI applications." Liu Weiguang, senior vice president of Alibaba Cloud Intelligence Group and president of the public cloud business department, said.

It is worth noting that during the conference, Alibaba Cloud released the industry's first full-lifecycle MCP service, aiming to promote the in-depth application of AI in all walks of life.

After listening to the meeting, Haifeng looked at science and technology, and sorted out the following highlights of the meeting for everyone.

The Beijing Municipal Government is helping the AI industry develop

As the host of this year's Alibaba Cloud AI Potential Energy Conference, Yang Shuo, Deputy Secretary-General of the Beijing Municipal Government, revealed Beijing's outstanding achievements in the field of artificial intelligence.

目前，在京人工智能企业已超2400家，核心产业营收超3000亿元，备案123款大模型产品，数量位居全国第一；数据基础制度先行区发布100个大模型高质量数据集，数据总量达150PB；智能算力规模超2.2万P，京津冀蒙环京算力供给廊道已形成。

In the face of the new pattern of industry competition brought about by the rise of emerging large models such as DeepSeek, how to further accelerate the in-depth application of large models in the industry has become a key issue. In response to this question, Yang Shuo gave three clear directions.

The first is to support the agglomeration and development of MaaS platforms in Beijing, support the cultivation of a number of representative demonstration projects and benchmark products through "hash power vouchers", and support a number of "first open" datasets through the "data voucher" policy;

The second is to promote industry software enterprises to fully possess model capabilities, and give priority to organizing and planning key projects in the fields of basic software, industrial software, transaction processing software, and new security software;

The third is to support software enterprises to accelerate the transformation of intelligent technology, promote enterprises to improve R&D efficiency through computing power construction, model deployment, data governance, etc., and give priority to organizing and planning key projects in basic software, industrial software, and other fields;

Fourth, improve the AI application service ecosystem, promote the construction of an international open source community for artificial intelligence, the implementation of open source platforms, and the construction of a shared ecosystem; Include new products and services empowered by artificial intelligence in the scope of support for small and medium-sized enterprises.

Alibaba Cloud continues to invest in promoting the implementation of AI

截至2025年1月底，阿里云百炼平台中调用通义API的企业和开发者已超过29万。AI已广泛应用于重构代码开发、硬件制造、智能座舱、金融服务、药物研发、太空探索等诸多场景，行业化应用经验正迅速积累与沉淀。

According to Liu Weiguang, senior vice president of Alibaba Cloud Intelligence Group and president of the Public Cloud Business Division, Tongyi Lingcode has been able to independently perform the whole development process tasks such as task disassembly, code writing, defect repair, and testing, and the development efficiency has been improved dozens of times.

目前，通义灵码插件下载量超1300万，累计生成超20亿行代码，为建设银行、中信证券、平安集团、中国人寿等上万家企业提供服务，成为国内最受欢迎的辅助编程工具之一。

The industry's first full-lifecycle MCP service was launched

At this conference, Alibaba Cloud Bailian officially launched the industry's first full-lifecycle MCP (Managed Compute Platform) service. In the past, when building an agent, enterprises needed to spend a lot of energy and resources on resource management, development and deployment, and engineering O&M, which was a cumbersome and complex process.

Everyone is very concerned, what improvements can Alibaba Cloud Bailian MCP service bring?

The author learned that it takes 50 minutes for users to quickly set up an agent that connects to MCP services. The platform integrates Alibaba Cloud Function Compute and has more than 0 industry-leading large models and more than 0 mainstream MCP services, comprehensively solving the problems of computing power resources, large model resources, and application toolchains required for agent development.

目前，百炼平台首批已上线包括高德、无影、Fetch、Notion等在内的50多款阿里巴巴集团及三方MCP服务，充分满足不同场景下的Agent应用开发需求。

Help Horizon train end-to-end intelligent driving models

The author pays attention to the fact that in the field of intelligent driving, Alibaba Cloud has also shown strong technical support capabilities.

As the largest mass-produced intelligent driving technology company in China, Yu Kai, founder & CEO of Horizon, said at the conference that relying on Alibaba Cloud's Wanka intelligent computing cluster, Horizon is carrying out end-to-end intelligent driving model training.

During the training process, through timely investigation of abnormal nodes and fine management of each training task, the overall fault recovery time of the intelligent computing cluster was reduced by more than 40%, which greatly promoted the acceleration of end-to-end model training efficiency.

The author has learned that Horizon has released the Journey series of on-board computing solutions and HSD full-scene intelligent driving solutions. Among them, the journey 2025P, which has the industry's top computing performance, has been officially returned to the film, and the actual vehicle verification has been completed within 0 hours. The HSD based on Journey 0P will also be delivered as the first mass-produced cooperative model in the third quarter of 0.

It is worth mentioning that Alibaba Cloud released the AI ecological partner program - "Flowers" plan, and launched the AI application and service market. In the next three years, Alibaba Cloud will work with ecosystem partners to serve millions of cloud customers, expand business opportunities of tens of billions of yuan, and jointly build a prosperous AI ecosystem in China.

Postscript: China's AI companies are accelerating their rise

When it comes to the conference held by Alibaba, the Apsara Conference is more well-known to the public. The Apsara Conference is a comprehensive technology feast, with cloud computing as the core, and extending to big data, artificial intelligence, Internet of Things, cloud native and other fields.

The Alibaba Cloud AI Potential Energy Conference is the first to be unveiled, which mainly focuses on the AI field and has three main values: one is to accelerate AI applications, the second is to explore the enterprise innovation paradigm in the AI era, and the third is to build a prosperous AI application ecosystem.

Judging from the many signals released by this conference, Alibaba Cloud is continuing to make efforts in the field of AI, whether it is the explosive growth of the number of Tongyi large model service enterprises or the innovative launch of MCP services, all of which show its strong technical strength.

Taking the Tongyi model as an example, it competes with the ByteDance Doubao model, Huawei's Pangu model, and Baidu Wenxin Yiyan, and each has different advantages and disadvantages.

Among them, the Tongyi model is deeply integrated with the Alibaba Cloud ecosystem, which can quickly access business scenarios such as e-commerce, customer service, and intelligent hardware, and performs well in tasks such as customer service systems, meeting minutes, and content summaries. However, the multimodal generation capabilities of the general large model, such as image generation and video understanding, are relatively limited, and the ability to process complex professional knowledge needs to be improved.

值得一提的是，4月2日，阿里通义千问近期开源的端到端全模态大模型Qwen2.5-Omni登上总榜榜首，非常争气。据悉，阿里至今已向全球开源200多款模型，千问衍生模型数量已突破10万，超越美国Llama系列，成为全球最大的开源模型族群。

Let's take a look at the development of HUAWEI CLOUD's Pangu model. It is reported that the Pangu model has been implemented in more than 0 industries and more than 0 scenarios, and has played an important role in many fields such as autonomous driving and industrial design. At present, the Pangu model has been updated to version 0.0, which contains models with different parameter specifications (billion-level, ten-billion-level, billion-level, trillion-level parameters) to adapt to different business scenarios. It can accurately understand text, pictures, videos, radar, infrared, remote sensing and other modalities; It has strong mathematical skills, complex task planning skills, and tool calling skills.

百度云的模型发展更是快速，文心一言很出名，近期不断更新。截至到去年11月，文心大模型日均调用量已超过15亿，文心一言用户规模达到4.3亿。文心一言依托百度强大品牌影响力和海量数据积累，在数据训练和用户信任上具有先天优势。因此，文心一言的语言理解和生成能力强，在知识问答、文本创作、逻辑推理等方面表现优秀；融合多领域知识，在不同行业应用中可提供丰富准确信息。

With the rise of domestic large models, competing Chinese AI companies are competing with international giants such as ChatGPT with a "hundred flowers blooming", and domestic large models have natural advantages in the Chinese context, industry know-how and policy adaptation.

The policy is blowing strongly, the computing power is surging, and the scene is sparkling. We have reason to believe that Chinese AI companies will lead the next golden decade of the global intelligent revolution with independent innovation.