Alibaba Cloud has dropped another bombshell

Author | Chai Xuchen

Editor | Zhang Xiaoling, Zhou Zhiyu

The industry has been eagerly awaiting DeepSeek-R2, but Alibaba Cloud has thrown out a bombshell first.

In the early hours of April 29, Alibaba Cloud officially launched the Qwen3 series model "full package," which includes 2 MoE models and 6 dense models.

With eight models ranging from 0.6B to 235B, they can be used on mobile phones and also provide a flagship experience, and they are open-sourced all at once, demonstrating great sincerity. Moreover, the small model Qwen3-4B can compete with the previous generation QwQ-32B. Its popularity is almost self-evident. Within two hours of release, the Qwen3 series had already surpassed 16.9k stars on GitHub.

Alibaba Cloud, which had been quietly accumulating AI + cloud capabilities, has seized the opportunity to ride the fast track of artificial intelligence. It has transformed itself from a cost center of the group over a decade ago into a true driving engine, injecting AI DNA into the entire Alibaba Group and breaking through growth bottlenecks; at the same time, it is preparing for the AI transformation of various industries with a cloud computing infrastructure + large model MaaS middle layer + application top layer architecture.

The market lacks the demand to break through the growth ceiling and the competitive situation; this AI revolution will soon trigger a surge in computing power demand, ultimately converting Alibaba Cloud into a continuous cash flow.

In this process, Alibaba's e-commerce empire will quickly stealth toward becoming an AI empire. As Jack Ma said internally: "AI is not a multiple-choice question, but a must-answer question for Alibaba's rebirth."

Midnight Shock

In the early hours of April 29, Alibaba's AI move Qwen3 finally launched just before the May Day holiday.

The new generation Qwen3 directly released 8 models of different architectures and sizes in a "full package," ranging from 0.6B to 235B, showcasing a powerful lineup.

This set of models has many highlights, with performance being the foremost. In benchmark tests for code, mathematics, and general capabilities, Qwen3 can compete with the current top models like o3-mini, Grok-3, and Gemini-2.5-Pro.

Qwen3 has bridged the generational gap between Chinese and American large models, but more importantly for the large model industry, Qwen3 integrates "fast thinking" and "slow thinking," becoming the world's first open-source hybrid reasoning model.

Earlier this year, DeepSeek-R1 showcased the magic of reasoning models to all users, thinking in a sequence like humans, continuously verifying, and ultimately deducing suitable answers. However, this repetitive "deep thinking" is clearly excessive for solving basic scenarios like checking the weather or asking common knowledge, requiring users to manually switch the "deep thinking" toggle based on their needs.

Qwen3 can choose "fast or slow thinking" like a human, with the model automatically recognizing task scenarios and seamlessly switching thinking modes. Zhou Jingren, CTO of Alibaba Cloud and head of Tongyi Laboratory, pointed out that such hybrid reasoning models will be an important trend in the development of the model industry The surprises brought by Qwen3 go far beyond this; in addition to its "explosive" performance and diverse functions, its deployment costs have also seen a dramatic decline.

This year, the Qwen team further recognized the existence of opportunities in areas such as intelligent driving and AI hardware. To better fit these scenarios, Qwen3 adopted a "multi-size" strategy and further achieved significant cost savings in deployment with its low energy consumption ratio of "small effort yields miraculous results."

According to publicly available data from the Qwen team, deploying the flagship version of Qwen3 requires about 4 H20s, whereas the fully powered DeepSeek-R1 requires 16 H20s to operate. Qwen3 has significantly lowered the cost of model adoption.

Various actions by Alibaba Cloud also confirm the "Jevons Paradox": when technological advancements improve efficiency, resource consumption does not decrease but rather surges. For example, Watt's improved steam engine made coal combustion more efficient, resulting in a spike in coal demand. This means that the extreme optimization of training costs by Qwen3 will stimulate more applications, services, and hardware to actively choose "AI transformation."

Thus, Alibaba Cloud's model open-source strategy is paving a clear path—becoming the foundational soil to accelerate the explosive growth of AI applications and continuing to write the narrative of China's technological leadership.

Taking Over DeepSeek

In the past two years, "thunder from a clear sky" in the AI field has almost become a daily occurrence. During the Spring Festival of 2025, DeepSeek ignited social circles, but its ripple effects are profound.

DeepSeek proved that large models can be "cheap and abundant" under the Chinese solution, bringing new hope for the rapid implementation of AI applications.

After the release of DeepSeek-R1, Baidu quietly adjusted to an open-source strategy, simultaneously integrating DeepSeek to strengthen its foundational search business; Tencent reversed its cautious attitude, fully embracing DeepSeek's integration into WeChat and vigorously promoting its own Yuanbao App; the automotive and industrial circles also rushed to integrate, prominently advertising the incorporation of DeepSeek into vehicles as a selling point.

It is worth mentioning that Alibaba Cloud also announced on the sixth day of the Lunar New Year that DeepSeek was available for deployment, supporting users to deploy it with one click on the cloud.

However, an insider from Alibaba Cloud pointed out that after DeepSeek was put on the shelf, the usage of the Qwen series dropped by nearly 20%. In response, Zhou Jingren did not directly address this but stated that it is normal for developers to use different models, and the open-source community should not be exclusive. Sources close to Alibaba Cloud indicated that the company's strategy is to attract users with a rich array of models and services, ultimately monetizing through the cloud.

Nevertheless, the popularity of DeepSeek continues to exert pressure on the Qwen team. How to maintain the position of open-source and uphold the leading edge of self-developed large models has become a new requirement for the company towards the Qwen team. To win this competition, the fully evolved Qwen3 was thus born.

The late-night release of a "nuclear bomb" reflects Alibaba Cloud's urgency to energize the industry and accelerate AI penetration. For Alibaba, this is a key counterattack after DeepSeek ignited domestic users' awareness of AI, leveraging the momentum for a leap forward At present, the significant enhancement of model inference capabilities has made Agents the hottest AI investment direction for 2025. In this wave of enthusiasm, Manus has become the first Agent to go viral domestically, marking the beginning of the Agent era.

What is an Agent? Simply put, while a chatbot like ChatGPT requires multiple instructions to complete a task, an Agent like Manus only needs the user's initial prompt to carry out a complex task from start to finish.

The highly popular Manus is essentially a "repackaging" based on the Qwen model.

To incubate more Manus and allow more ordinary users to experience the "disruption" brought by AI applications, Qwen3 has further enhanced its native Agent framework against the backdrop of significant capability improvements and cost-effectiveness, supporting the currently popular MCP protocol, aiming to enable everyone to use Agents.

At last month's Alibaba Cloud AI Momentum Conference, Alibaba Cloud previewed the ecological vision of the Agent Store. By opening up the Agents of Alibaba Group and its ecological partners through the Agent Store, people from various industries can have their own dedicated assistants. At this time, the Tongyi large model still serves as the foundation, but "invisibly" supports users and clients in the form of Agent applications.

In the past, Apple relied on the App Store to drive the iteration from feature phones to smartphones, and Alibaba Cloud aims to leverage its AI ecosystem to accelerate enterprises, developers, and users into the era of artificial intelligence.

The widespread application of AI products will further promote the use of cloud computing resources.

Over the past two and a half years, from the generative AI computing brought by ChatGPT to the Agent computing represented by Manus, and to the future embodied intelligence and intelligent driving that AI will bring to the physical world, each migration and maturation of AI technology will lead to an exponential increase in computing demand.

Currently, Qwen3, with deployment costs less than 40% of DeepSeek R1, has comprehensively outperformed leading domestic models, further lowering the cost of model proliferation. High-performance, cost-effective models will accelerate the Cambrian explosion of AI applications in China.

Alibaba Cloud predicts that 2025 will be the inaugural year for AI applications, especially AI Agents, with the demand for computing power growing exponentially. Internal data shows that an AI product with a daily active user count in the millions, if transformed into an Agent, would consume tokens 300,000 times more. This represents a continuous cash flow for Alibaba Cloud.

Comprehensive AI Integration

In the past two years, Jack Ma has frequently emphasized AI during his public appearances. At the end of last year, he reflected that 20 years ago, they were fortunate to seize the internet, which led to everything they have today. He firmly believes that AI will bring about a significant transformation.

"Rebuilding Alibaba with AI" has become an internal consensus. At this moment, Alibaba Cloud is embedding AI into the DNA of the entire group, transforming itself from a cost center of the group over a decade ago into a true driving engine A grand leap in AI at Alibaba has officially begun.

Since proposing and implementing the "AI-driven" strategy two years ago, Alibaba has continuously increased its investment in AI infrastructure. Earlier this year, Eddie Wu announced that the group would continue to ramp up its efforts, spending 38 billion yuan to build cloud and AI hardware infrastructure over the next three years.

This significant amount not only sets a record for similar investments by Chinese companies but is also nearly equivalent to Alibaba's cumulative profits over the past three years. This means that Alibaba is betting the money it has earned over three years on AI.

With an almost all-in momentum, Alibaba Cloud has fully reconstructed a full-stack technology architecture system, covering everything from underlying hardware to computing, storage, networking, data processing, model training, and inference platforms, becoming one of the earliest and most deeply laid out AI infrastructures developed in-house in China.

With the support of this infrastructure, Alibaba will first use AI to "transform" itself. As Alibaba possesses the richest commercial ecosystem in China, it inherently has practical scenarios for AI implementation.

The group quickly initiated a comprehensive goal alignment across its entire system. According to insiders, the performance evaluation of all business units at Alibaba for 2025 will be directly linked to the effectiveness of AI technology applications. Among them, core e-commerce departments such as Taobao and Tmall are required to fully integrate AI capabilities, with over 200 business lines currently collaborating with the Tongyi Qianwen team to develop intelligent features.

At the forefront is the entry point for consumer spending, which is also the foundation of the group. According to Eddie Wu, through AI technology transformation, Taobao can significantly enhance interaction and transaction efficiency with consumers, while also having vast potential in many consumption areas such as shopping decisions, leading to an increase in user engagement and value, and is expected to bring greater user value beyond shopping.

Just like when Alibaba was founded, leveraging advanced productivity like the internet to gain "unlimited resources." In the future, whether it is delivering a better user experience, achieving higher operational efficiency, or enabling every party in this commercial system—users, merchants, and the platform—to gain greater benefits, new productivity increments will be necessary, and AI is most likely to be the weapon that breaks the productivity bottleneck.

Next, another important direction is to strengthen the consumer end, with Tongyi Qianwen and Quark, closest to AI transformation, being pushed to the forefront.

Quark can be considered the first wave of AI products from Alibaba to reach the public. The recently launched "AI Super Box" has already shown users its embryonic form as a super productivity tool and agent.

Over the past year, Quark has accelerated iterations around tens of millions of young internet users, making its ambition to become a "national-level AI tool" evident. In February of this year, Jack Ma specifically visited the Quark office area. A month later, Quark was revamped and upgraded, rising to the status of Alibaba Group's flagship AI application, becoming a hope for Alibaba in the ToC application direction.

On the other hand, after the launch of Qwen3, the Tongyi APP, as the "favorite child," quickly integrated three models of Qwen3: A22B, 32B, and the high-performance deep thinking version of 32B. Clearly, in the context of ByteDance's Doubao and Tencent's Yuanbao making significant efforts in the consumer end, Alibaba also wants to challenge its original "soft spot," attempting to replicate the breakout capabilities similar to DeepSeek through AI Immediately following, the combination of "Tongyi + Quark" began to be rapidly replicated within Alibaba—In late April, Amap launched an AI navigation intelligent agent, Fliggy introduced the "AI Q&A" feature, and Ele.me launched the rider AI assistant "Xiao E."

It can be said that Alibaba has fully aligned its tasks and goals internally, using AI to initiate a grand march to enhance productivity. From cloud computing, core e-commerce to global business expansion, Alibaba has already transformed itself into an AI company under the "AI-driven" strategy.

Next, the explosion of AI applications is about to drive a revaluation of Alibaba.

Welcoming Revaluation

In terms of current layout, Alibaba is the most resolute and comprehensive player in AI investment among Chinese internet companies.

In the next three years, a capital expenditure of 380 billion on AI is unprecedented; Alibaba has a complete layout in infrastructure, large models, and applications, with Alibaba Cloud and the Tongyi large model family both ranking in the global first tier.

This "cloud + AI" snowball growth will raise the ceiling of the "revaluation narrative." Since the Cloud Summit last year, Alibaba Cloud has made its intentions clear—computing power upgrades, large model iterations, and significant price reductions.

Currently, Alibaba's large model capabilities are globally leading. Since the global surge of ChatGPT, Alibaba has maintained a high frequency of model updates, continuously catching up with the world's leading large models. The emergence of Qwen3 represents that the generational gap between Chinese and American large models has been bridged, and it has even taken the lead in the open-source field ahead of American large models.

In terms of infrastructure, Alibaba Cloud is the leading cloud computing platform in the Asia-Pacific region and globally. With the accelerated development of artificial intelligence, the classic computing system centered on CPU is rapidly shifting to a GPU-dominated accelerated computing system. Alibaba Cloud, which once led in CPU computing power, now needs to "switch tracks" to bind with AI players.

Currently, under the attraction of the Qwen model family and the open-source ecosystem, over 90% of listed commercial banks, mobile brands, automotive brands, and leading home appliance companies have joined Alibaba's AI ecosystem, and even Apple has chosen Alibaba as an important partner for its AI layout in China.

Facing the AI era, Alibaba Cloud has fully reconstructed a full-stack technical architecture system from underlying hardware to computing, storage, networking, data processing, model training, and inference platforms, becoming one of the earliest self-developed and deeply laid out AI infrastructures in the country.

At the application level, Alibaba has the richest commercial ecosystem in China. Quark is the application with the highest user volume in AI search in China. DingTalk is one of the few SaaS platforms in the B2B field in China that possesses both scale and ecological advantages. Alibaba's e-commerce, map navigation, content, and other business sectors are also actively applying AI.

Thus, Alibaba has formed a structure with cloud computing as the underlying layer, large model MaaS as the middle layer, and applications as the top layer, transforming itself into an all-round player, making it the most resolute and comprehensive player in AI investment among Chinese internet companies.

It should be noted that OpenAI has been leading the pack because it occupies an absolute leading position in both self-developed models and AI applications, and the combined economic value is all reflected in its overwhelming valuation compared to peers. Now, this valuation model will also be applied to Alibaba Alibaba's AI empire is gradually taking shape, and its business revenue will soon become a barometer of its ambitions and market enthusiasm.

On April 30, IDC released the report "China Public Cloud Service Market Tracking," which shows that AI is driving a recovery in the public cloud market. In the second half of 2024, the scale of China's public cloud IaaS market is expected to reach 94.82 billion yuan, a year-on-year increase of 13.8%. Alibaba Cloud ranks first with a market share of 26.1%, marking a continuous rebound for three consecutive quarters.

Alibaba Cloud's quarterly market share increased from 25.8% in Q1 of last year to 26.2% in Q4. As market share rises, Alibaba Cloud's profitability also improves significantly.

According to Alibaba's financial report, Alibaba Cloud achieved revenue of 31.742 billion yuan in the fourth quarter of 2024, a year-on-year increase of 13%, nearly doubling compared to the previous quarter's growth rate of 7%. Thanks to the growth in public cloud revenue driven by AI-related products, AI revenue has seen triple-digit growth for six consecutive quarters.

From an internet e-commerce company to an AI technology giant, Alibaba Cloud is racing forward with the entire Alibaba Group on the path of revaluation.

Goldman Sachs analyst team led by Ronald Keung predicts that Alibaba Cloud's AI-related revenue will reach 29 billion and 53 billion yuan in fiscal years 2026 and 2027, accounting for 20% and 29% of total revenue, respectively. This will further drive Alibaba Cloud to achieve over 20% revenue growth in fiscal years 2026-2027.

JP Morgan estimates that for every 2 percentage points increase in cloud revenue growth due to AI, Alibaba's stock price will rise by 1%. In their view, if calculated at Microsoft's valuation multiple of 10.5 times, Alibaba Cloud's value could reach as high as $185 billion, and Alibaba's corresponding market value would reach $391 billion. Currently, Alibaba's market value is $304.5 billion.

Reinventing Alibaba with AI is a crucial juncture in the revaluation of this giant. Perhaps the breakthrough of Qwen3 is just the beginning of this grand narrative.

The AI race is far from over, and Tencent Cloud and Huawei Cloud are also accelerating their follow-up in areas such as intelligent driving data and cloud computing. For Alibaba, 380 billion may only be a ticket to entry; the real test lies in whether it can continuously convert technological advantages into commercial momentum and tangible profits.

To ultimately become the "water, electricity, and coal" of the AI transformation and build a prosperous new generation of AI, Alibaba still needs to accelerate its pace