Recommendation
Alibaba's AI strategy has quietly begun, leading the industry to change.
Core content:
1. DeepSeek has set off a revolution in the large model industry, and domestic manufacturers have adjusted their strategies
2. Alibaba's AI strategy is clearly visible, and the entire system has been aligned to the goal even though it has not been released
3. AGI has become Alibaba's primary goal, creating Asia's largest AI infrastructure and best practices
Yang Fangxian
Founder of 53AI/Most Valuable Expert of Tencent Cloud (TVP)
The entire Chinese technology community probably didn’t have a peaceful Spring Festival in 2025. DeepSeek overturned the table of the big model industry in a way that no one expected.
Although this happened several months ago, the chain reaction of this incident is far-reaching. The public and the media are more concerned about the changes in startups, such as the "AI dragons", but in fact, the impact on domestic "big companies" is equally huge.For example, Baidu began to quietly adjust its strategy to open source models, canceled the subscription fee of Wenxin, and integrated DeepSeek-R1 to further develop its core business of "search"; Tencent, contrary to its Buddhist and cautious approach, deeply integrated DeepSeek-R1 into its entire application family including WeChat; for example, ByteDance is said to have started deep reflection on how to effectively balance the efficiency of technological improvement and the support for business needs after it has made up its mind to "all in big models".Over the past two years, the AI field has been changing rapidly. I remember that at the NVIDIA Appreciation Dinner held in Beijing earlier this year, Manus Chief Scientist Ji Yichao (Peak) asked Huang Renxun what would happen in the AI field that would surprise him. Huang's answer was "nothing" - this is an unpredictable industry stage.Small companies compete on agility and speed, so the impact of this wave of shocks is actually not that bad. But for large companies, I think the challenge may be greater, because no matter whether you have a large "strategy department" or not, how to set strategy for a large organization in an unpredictable stage of industrial development is bound to be a very "sour and refreshing" challenge.In the past few months, domestic large companies have gradually released their AI strategies in such occasions as earnings conferences, but most of them are relatively fragmented. I recently focused on Alibaba and found that its strategic fragments have been effectively "pieced together" into a clearer picture, and some strategic intentions are "coming to light".Objectively speaking, Alibaba has never really "released" its AI strategy. But I think Alibaba may be the first giant in China to form a clear three-year AI plan and 10-year goals. More importantly, it seems that Alibaba has quietly aligned its goals and tasks across the entire system and started the march of the army.The goals and actions of this "big company" in the AI era will inevitably have an impact on the industry. Today, let's discuss the "Ali Strategy" that I put together.Let me give you a simple outline. If we use the jargon of “mission, vision, and values” to define Alibaba’s “goals, results, and methods” in the AI era, it would be roughly:AGI must be Alibaba’s top priority
Alibaba must become Asia's largest AI infrastructure and best AI practice provider
Open source! Open source! Open source!
AGI, why must it be Alibaba
The most important strategic goal in the AI eraMany people probably saw that Alibaba returned to high growth in the financial report season at the beginning of this year. Alibaba Cloud's AI-related products have maintained a three-digit percentage growth rate for six consecutive quarters. At this financial report meeting, CEO Wu Yongming systematically talked for the first time about what opportunities Alibaba should seize in the AI era based on its own capabilities and talents.Let's first look at Alibaba's current status in the AI industry: At present, Alibaba Cloud, Qwen model family and its open source ecosystem, plus many ToC application scenarios such as Taotian and Quark, make Alibaba at least the most formidable AI player in the Asian market. This must be the starting point for Alibaba to judge its own AI opportunities at the moment, and also an opportunity for the next three or even ten years.On the other hand, Alibaba's extension of "making business easy in the world" in the AI era must provide infrastructure for AI innovation and transformation in thousands of industries. This means that every layer of platform opportunities from computing power to models to applications - Alibaba Cloud, Qwen model family and its open source ecosystem, and application platform need to continue to evolve.At this earnings conference, Wu Yongming made clear the most important task between the cards in hand and future industry goals - his original words were: "The first goal must be to pursue the realization of AGI and continuously pursue breakthroughs in the boundaries of the model's intelligent capabilities."It is not new for a company of Alibaba’s level to say that it will “strive to pursue AGI”, and they are indeed qualified to say so. However, how serious Alibaba is in saying this may still depend on the actual significance of this matter to Alibaba.In fact, after the emergence of this wave of AI, the psychological projection of this technical variable on domestic large companies may be different. Today, I will not talk about the "psychological state" of ByteDance and Tencent. Anyway, I think that for Alibaba, AGI is the key direction for Alibaba's business to "turn from defense to offense". This is why it must resolutely take AGI as its first goal, and there is no way back.Over the years, with the rise of Pinduoduo and Douyin, Alibaba's e-commerce position has been "attacked", and it is clearly on the defensive. Another important business, cloud computing, is similar. Alibaba Cloud has also taken a detour. The growth rate has been in the single digit for many years, which has given more forces such as Volcano Engine and Baidu Smart Cloud room to "cut the meat with a knife".What does "guarding" mean? What does "cutting meat with a knife" mean? No one knows this passive feeling better than Alibaba. On November 29, 2023, Alibaba's market value was about to be overtaken by Pinduoduo. The posts circulating on Alibaba's internal network were full of the loneliness, helplessness and unwillingness of the once proud young people. It was also on that night that Alibaba's founder Jack Ma rarely said, "I firmly believe that Alibaba will change and Alibaba will change."His next sentence was that the era of AI e-commerce has just begun, and it is both an opportunity and a challenge for everyone.If you can’t defend, how can you attack? At this point in time, the times gave Alibaba a new variable - AGI. To be more specific, general artificial intelligence is a fundamental upgrade of productivity. All business competition, industrial structure, profit and business model are ultimately "production relations", and the fundamental change of productivity is the most important breakthrough to change the original situation.Many people should remember that Alibaba co-founder and chairman of the board Joseph Tsai mentioned the internal reflection that "Alibaba has fallen behind in the past few years." He mentioned that the essence was to forget who the real customers were. For example, it did not provide the best experience for users who used the App to shop, and did not really focus on creating value for users.The so-called "forgetting" actually means "insufficient resources investment" in the corresponding goals. The truth in the business world is that an organization can keep pace with the times and innovate continuously, and keep leading the times. It is less likely to come from a god-level decision of "foresight" and more from long-term correct direction, continuous investment of sufficient resources, and even "meaningful waste".All companies have limited resources. In an environment where more and more forces are "cutting flesh with a knife", defending positions by fighting for every inch of land, passive response and reflexive pursuit cannot reverse the situation. This is a dead end of being divided and surrounded. Today, Alibaba needs to break through collectively, and it needs to return to the energy field when it was founded - just like using advanced productivity such as the Internet to obtain "unlimited resources" at that time. In the future, whether it is delivering a better user experience, bringing higher operational efficiency, or allowing users, merchants and platforms, everyone in this business system to obtain greater benefits, new productivity increments are needed, so only improving productivity can truly solve the problem.In the final analysis, the reason why "pursuing AGI" must be Alibaba's first goal in this era is that Alibaba realizes that the revolution it started is far from victory. It is necessary to cheer up, return to Alibaba's origin, and continue to use "advanced productivity" to carry out the revolution to the end.Because only by being a revolutionary can you avoid being revolutionized.What goals should be achieved during the three-year AI infrastructure period?After sorting out the big logic, let's take a look at what clear goals Alibaba has. At present, the next three years will be an important period for Alibaba's AI infrastructure construction.In combination with Alibaba's business characteristics and AI strategy, Wu Yongming also gave a further strategic map at the earnings conference in February this year. The most clear statement is that in the next three years, Alibaba will increase investment in three major areas:Infrastructure construction for AI and cloud computing;
AI basic model platform and AI native applications;
AI transformation and upgrading of existing businesses.
Among these, AI infrastructure is the most certain opportunity at the moment. Wu Yongming gave a clear investment expectation - the next three years will be the three years with the largest and most concentrated cloud computing and AI infrastructure construction cycle in the history of Alibaba Group. It will invest more than 380 billion yuan to build cloud and AI hardware infrastructure, and the total amount will exceed the total of the past decade.This clear investment expectation is based on the industry signal seen behind the three-digit year-on-year growth in revenue of Alibaba Cloud's AI-related products for six consecutive quarters. Customer demand continues to grow, and the growth rate exceeds the original judgment. Especially after the Spring Festival this year, DeepSeek-R1 has led to a surge in the demand for AI reasoning in the entire industry. More than 60%-70% of Alibaba Cloud's new customer demand is for reasoning.In fact, looking at the long term, when AI expands from training to the inference stage, the demand space and opportunities for AI cloud computing have just begun.In the past two and a half years, from the generative AI computing brought by ChatGPT, to the agent computing represented by Deep Research and Manus, to the embodied intelligence and autonomous driving that will change the physical world in the future, every migration and maturity of AI technology will cause the amount of computing to grow exponentially.Without going into the distant future, the inference cost of asking ChatGPT once is almost zero, but asking Manus once costs 2 US dollars, because the latter will continuously use the previous token as the input context when generating the next token, and perform perception, planning, and action step by step. Behind each question are dozens or hundreds of inference model calls and calculations.The new opportunities and space brought by AI reasoning also mean that now is the window period for AI cloud computing to undergo a huge change, which will promote the optimization and upgrade of the supply-side Infra layer computing architecture. In other words, the existing computing solutions of the previous era may not be the best choice in the new era, and there is huge room for innovation.For example, the achievements of DeepSeek in Infra are as significant as the release of DeepSeek-R1. DeepSeek's approach puts forward the goal of optimizing Infra in the AI era: through software and hardware integration optimization, rather than buying the most expensive hardware, AI computing can achieve the ultimate utilization rate to increase token throughput per unit time and reduce computing latency. This also determines the reduction of inference costs and the further penetration of AI applications.Two years ago, when I visited SambaNova in Silicon Valley, I heard founder and CEO Rodrigo Liang talk about the need to meet the computing needs of the AI era through an integrated hardware and software design. At that time, the optimization of inference computing was not as urgent as it is now. Now SambaNova, including other cloud vendors, chip companies, inference algorithm companies, and model companies are all working on AI Infra. Amid the upsurge, a group of new companies such as Fireworks, Groq, and Cerabras are winning the adoption of more and more AI application developers.The same is true in China. AI Infra companies have been the hottest startup projects in the past two years. Chips, interconnections, inference algorithms, heterogeneous computing clusters, etc. are all experiencing new opportunities for upgrading under the new AI technology.A giant like Alibaba may not be as agile as a startup in the midst of the dramatic changes in AI Infra, so it should do its part as a big company - when everything is unknown and uncertain, investment is the most certain thing. The ToB gene of Asia's largest cloud vendor will also give Alibaba more advantages in exploring cloud computing in the AI era. In my opinion, this is the logic behind Alibaba's strategic determination to invest 380 billion in three years.Of course, for this investment in the concentrated construction cycle of the next three years, Alibaba has most likely seen a clear path to simultaneously transform and promote its business.First: Let Agent capabilities be “nailed” into the industry.For SaaS and enterprise internal software customers of Alibaba Cloud services, they are undergoing a transition from providing tools to providing productivity. AI Agent allows these software companies to directly deliver reliable results. In the foreseeable future, many internal systems of enterprises will gradually become interconnected and called by a considerable number of AI Agents to help enterprises improve efficiency and even help enterprises complete some very important decision-making work. There are quite a few opportunities for upgrading various aspects of SaaS software and the PaaS behind it.At the same time, in the AI era, DingTalk can become a new entry and interface in the entire enterprise software market and enterprise collaboration market, forming effective collaboration with other software ecosystems or AI agents, which will further increase the demand for cloud computing. Of course, the premise of this is that DingTalk needs to more sharply promote the product transformation from "providing tools to outputting productivity."It can be predicted that in the future, DingTalk will need to refocus on its products. Driving the value of the cloud is not the goal, but the result of making a good product.Second, seek the value of ecosystem and cloud capabilities by accelerating model technology and establishing open source leadership.In fact, this growth path has been verified in Alibaba Cloud's past performance. Many entrepreneurs and developers in the Geek Park community shared with me that although the open source model seems to be unprofitable and only for technology branding, in fact, the Qwen series of open source models has brought Alibaba Cloud real revenue growth. It can be said that it has been Alibaba Cloud's best sales in the past year or so.The adoption rate of the Qwen open source model family is actually very high. Compared with a general model API, most companies will further develop based on the open source model combined with their own scenario data. The efficient and high-quality Qwen open source model is almost the most popular model among companies and developers. According to data from Hugging Face, the world's largest AI open source community, as of February 2025, the number of derivative models of Alibaba Cloud's Qwen open source model has exceeded 100,000, ranking first in the world's largest open source model.After choosing the Qwen open source model, which cloud should be deployed on? A natural choice is to purchase Alibaba Cloud, because the entire Tongyi and Tongyi's derivative models run most efficiently on Alibaba Cloud.At the same time, in the post-training stage, there will be a huge market space for model customization and adaptability to different industries and even different private data, which will create a lot of room for value creation. In the future, there will be a lot of professional models or vertical models hosted on the cloud, which is another benefit of the prosperity of the Qwen open source ecosystem.Now, with the iteration speed of open source models, the differences between everyone on the base large model are actually gradually getting smaller, and open source models are becoming more and more available. In addition, as inference models based on OpenAI o1 or DeepSeek-R1, QwQ-32B become mainstream, more and more cloud computing will be needed. Because whether open source or closed source, most models will eventually be hosted on the cloud. Because the amount of calculation required by such a huge model can only be achieved with the highest efficiency on the cloud computing network. Through data centers all over the world, it can be delivered to application developers around the world in a faster way.Third, MaaS will become a very important component of the cloud business model.Although most enterprise customers will do post-training and secondary development based on the Qwen open source model, many developers will choose "model as a service", Tongyi API. Judging from the growth of Alibaba Cloud in the past six quarters, customers using Tongyi API will drive the use of quite a lot of other cloud products, which is a very obvious effect of customer-related sales. So to some extent, although Tongyi may not directly bring in very large revenue, it is a very important component of the cloud business model. With the continuous improvement of model capabilities and the increase in future customer demand, API charges will still exist as the most basic way to monetize the model.Therefore, for Alibaba, no matter how model capabilities and AI applications evolve in the future, AI and cloud computing infrastructure have a very clear business model, which is the cloud computing network. If artificial intelligence is the biggest commodity in the future, analogous to electricity, then the cloud computing network is analogous to the current power grid.At the same time, in the AI era, cloud computing is still a business that competes on scale effect and network effect. Especially at this stage, scale effect is very important. The above three clear paths will drive Alibaba Cloud to continue to expand in scale, which will bring more customers or customer usage scenarios on the one hand, and improve the overall investment cost on the other.To change the world, change yourself first:Alibaba needs to set an example in the AI eraAI basic big models and native applications are Alibaba's first strategic goal in the AI era, and AI infrastructure is a deterministic opportunity and investment. At present, Alibaba still has its own "cultivation" - Alibaba's e-commerce and other Internet platform businesses need to continue to shine as a role model in the face of new opportunities in the AI era.At present, the upgrade of AI technology has brought opportunities for Alibaba's many Internet platform businesses to significantly increase user value. From this perspective, Alibaba will continue to increase its investment in R&D and computing power for AI applications, use AI technology to deeply transform and upgrade various businesses, and seize new development opportunities in the AI era.In fact, within the Alibaba ecosystem, +AI has always been active. Even before the emergence of OpenAI o1 last year, when the industry was doubting whether "pre-training was slowing down and the limit of AI capabilities was just like this", various applications did not slow down the pace of +AI, but continued to infiltrate AI from the perspective of serving users well. Take DingTalk as an example. From the "slash" smart assistant in 2023 to the AI search and agent capabilities within the enterprise this year, it can obviously help me complete many tasks more smoothly.Based on the capabilities that AI technology can currently achieve, Alibaba currently has three clear directions for breakthroughs in AI applications:First of all, it is the entrance to daily consumption, which is also the foundation of Alibaba.In February this year, Alibaba's e-commerce business group CEO Jiang Fan stated at a financial report conference that domestic e-commerce needs to improve user experience and stickiness, and improve business efficiency. The medium- and long-term goal is a healthy and stable market share.In terms of these goals, according to internal sources at Alibaba, AI is beginning to play a role. Recently, with the transformation of AI technology, Taobao has made considerable progress in strengthening interaction with consumers and promoting transaction efficiency. Taobao also has many projects under development, which are expected to be gradually launched in 2025. The overall direction is that Alibaba believes that the application of AI technology in many consumer fields related to shopping decisions has great potential for increasing the value of Taobao users.The second "prominent" of Alibaba is Quark, which many people have not really observed seriously. This product can be said to be the first wave of AI products within Alibaba Group that rushed to the public. In the past year and a half, Quark has accelerated its iteration around tens of millions of the youngest Internet users. The goal of the product is likely to be "to upgrade itself to a national-level tool with AI capabilities."At present, the AI to C products defined within Alibaba Group are mainly the two apps, Quark and Tongyi. The Quark Business Unit was established in 2021 and was one of the most valued businesses in the Intelligent Information Business Group at that time. In 2023, under the AI-driven strategy, Quark was also promoted to Alibaba's first batch of strategic innovation businesses.At present, Quark has the largest number of users in China in the field of AI search. The application of AI big models here has great room for improvement in user search, productivity creation, and work efficiency. For example, the "AI Super Frame" recently launched by Quark allows users to see the prototype of Quark as a super productivity tool and agent, because it can be seen that personalized services are provided by integrating memory functions and various tool capabilities.As Quark's business opportunities become clearer, Quark has also gradually risen from an innovative business of the group to a strategic product. The evidence is that after the latest organizational adjustment in March this year, Wu Jia, President of Alibaba Group's Intelligent Information Department and Quark CEO, began to report directly to Alibaba Group CEO Wu Yongming, strengthening Quark's positioning as Alibaba's AI flagship application.Alibaba’s third AI ToC application is most likely AutoNavi, and its purpose should be to explore a new lifestyle of using AI to establish life services.As a product with more than 170 million DAU, AutoNavi is currently more of a navigation tool for users, but in the future it certainly hopes to become a portal to life services through the application of AI technology.Take the recently popular MCP (Model Context Protocol) as an example. This is a protocol launched by Anthropic in 2024. It is like Type C in the model world. It builds a bridge between AI models and various external data sources and tools, allowing different AI models, software tools and data sources to understand each other's intentions in real time and automatically form a task collaboration network.Many people don’t know that Amap’s API was the first in China to be fully compatible with the MCP protocol and launch the MCP Server. With this protocol, Amap integrated its own map services with intelligent algorithms, providing a simple and easy-to-use map service interface. After that, many companies and developers directly called Amap’s various location-based services through the MCP Server in various AI Agent applications.Of course, Taobao, Quark, and AutoNavi are just the first batch of Alibaba to see the opportunity of Internet platform products in the AI era. As AI big models have stronger and stronger intelligence capabilities, Alibaba's ToC platform opportunities may be very different. But there is no doubt that the ToC platform is also the strategic focus of Alibaba to continue to maintain its advantages in the new era of AI.In addition to Alibaba's own exploration, I think it is possible that it will also open up external channels for investment and acquisition. This will form two "new driving forces" in the industry together with ByteDance, which is currently quite active in China, and may also be a good thing for China's AI entrepreneurs.At this point, Alibaba’s first version of the “unreleased but currently being executed” overall strategy in the AI era has been fully presented. Objectively speaking, this strategy is not too surprising, and most of it is logically inevitable.But for Alibaba, a business trend from "passive response" to "active construction" is taking shape. Allowing a huge system to complete "unified thinking and look forward" as soon as possible may be the best posture to welcome a new era.
