At the intersection of technology and capital, a new storm is brewing. This time, the center of the storm is no longer the code and algorithms of the virtual world, but the bridge connecting AI to the physical world—embodied intelligence. Following this trend, AI leader Alibaba has also entered the fray.
Recently, Lin Junyang, the technical lead of Alibaba's Qwen, announced the formation of an embodied intelligence team within Qwen. This marks Alibaba's official entry into physical AI. Just a month earlier, Alibaba Cloud led a $140 million funding round for domestic robotics startup Agibot.
From investment to direct participation, Alibaba's ambitions are clear: it is going all-out to find a "body" for AI that can perceive and transform the physical world. Behind this lies a profound era backdrop: we are in a peculiar AI "fault line" era.
On one hand, AI's "brain"—represented by large language models like Qwen and GPT-4—is evolving at unprecedented speed, demonstrating remarkable cognitive and generative capabilities. On the other hand, AI's "body" appears clumsy. In the physical world, except for industrial robotic arms in specific scenarios, we can barely see intelligent physical entities capable of autonomous environmental understanding and human collaboration.
This fault line also contains enormous opportunities. Last year, Wu Yongming stated bluntly that AI's greatest potential lies in connecting to the real world. At this year's annual shareholders meeting in June, Jensen Huang emphasized that AI and robotics represent the two largest growth opportunities, worth trillions of dollars.
Currently, giants like OpenAI, Google, SoftBank, and NVIDIA are all heavily investing in the embodied intelligence track. A war to "find bodies for AI" has quietly begun. For Alibaba, which stands at a crossroads of transformation and value reassessment, this will also be a crucial pillar in achieving its grand vision of ASI and becoming "great again."
**From Cloud to Reality**
On October 8, Lin Junyang released a significant announcement: the team has officially established a "Robotics and Embodied Intelligence Group."
Lin pointed out on Twitter that multimodal foundation models are transforming into foundational agents that can utilize tools and memory, conducting long-horizon reasoning through reinforcement learning. "They should definitely move from the virtual world to the physical world."
This means Alibaba is determined to help Qwen transition from the language world to the physical world. At the recent Apsara Conference, Alibaba Group CEO Wu Yongming asserted that the industry is entering the "autonomous action" phase of AI. Wu said that in the future, there may be more intelligent agents and robots than the global population working and living alongside us.
For Alibaba, pushing AI from the cloud to the physical world is a crucial step in its grand blueprint toward artificial general intelligence (ASI). In fact, Alibaba has been laying groundwork for entering embodied intelligence for some time.
Since 2024, Alibaba Group has successively invested in a batch of embodied intelligence companies including Zhiju Dynamics, Star Era, Unitree Robotics, Starmap, and Ling Xinqiaoshou. Alibaba Cloud led Agibot's $140 million funding round in September.
Industry insiders indicate that since the first half of last year, Alibaba has focused heavily on embodied intelligence, "investing in almost all embodied intelligence companies." This is because robots are the ideal vehicle for large model capabilities to transition from the digital world to the physical world. Alibaba has secured positions in the track by investing in companies with different technical approaches like Unitree, Zhiju Dynamics, and Agibot.
Clearly, Alibaba's embodied intelligence layout is not a momentary impulse. To understand its motivation, it must be placed within Alibaba's current internal and external environment and its judgment of future technology endpoints.
First, this is strategic inevitability: AI's endpoint must be the physical world. As large language model capabilities mature, their further value realization necessarily requires the ability to perceive, understand, and transform the physical world. An AI that only runs in dialog boxes has a visible commercial ceiling, but an AI that can interact with the environment through robotic entities has unlimited potential.
Alibaba's moves are not isolated in the global AI landscape.
In the same week, SoftBank Group announced the acquisition of ABB's industrial robotics business for $5.4 billion; NVIDIA CEO Jensen Huang again mentioned that AI and robotics are "trillion-dollar long-term growth opportunities." Elon Musk views his humanoid robot "Optimus" as the core of Tesla's future value, first aiming to revolutionize his own factory production models.
Global giants are all entering the field, making physical AI the next battlefield. For Alibaba, the foundation of its commercial empire—e-commerce, logistics, and new retail—are all closely connected to the physical world. Whether it's Cainiao Network's global smart warehousing or Freshippo's automated supply chains, there are numerous scenarios requiring interaction with physical entities.
These scenarios serve as the most direct and expansive testing grounds for embodied intelligence applications. Therefore, "embodying" AI capabilities is an intrinsic need for upgrading its core businesses.
Facing such intense competitive landscape, Alibaba's absence would be tantamount to handing over future physical world gateways to others. This is Alibaba's urgent need to tell new growth stories for the capital market and inject a "stimulant."
In recent years, Alibaba's core e-commerce territory has faced challenges from new formats like PDD Holdings and Douyin. It urgently needs an imaginative new story to reignite investor confidence. Cloud computing and AI are undoubtedly the best protagonists of this new story, while embodied intelligence adds the most hardcore chapter.
It extends Alibaba's capabilities from online to offline, from virtual to real, sending a clear signal to capital markets: Alibaba is evolving into a more foundational technology platform company capable of defining the future shape of the physical world.
Industry researchers point out that through direct participation in robot development, Alibaba can both validate large models' adaptability in real physical environments, driving model evolution in noise processing and causal reasoning, while expanding data collection dimensions from text and images to mechanical control and sensor feedback applications.
This transformation from business model innovation to hardcore technological innovation is key to reshaping market valuation and launching a second growth curve.
**Bridging Industry Gaps**
"Now AI writes and paints better than 99.99% of people. But getting AI to actually work is still a wasteland," said Wang Xingxing, founder of Unitree Robotics, at a roundtable forum during the Bund Conference a month ago.
We are currently in a technological "fault zone." On one side, AI "brains"—large models—are explosively advancing at exponential speeds; on the other side, the clumsiness of AI "bodies" contrasts sharply with the complexity of the physical world. This gap limits AI's potential.
Players at the table have reached consensus: giving AI bodies and entering the physical world to "work" represents a vast blue ocean.
Morgan Stanley predicts that by 2050, 1 billion humanoid robots will be deployed globally, with a market size reaching $5 trillion. This figure is approximately double the total revenue of the world's 20 largest automakers in 2024. According to CICC estimates, China's humanoid robot potential market space could reach 22.8 trillion yuan by 2050, with a compound annual growth rate of 24.7% from 2024-50.
Such certainty has pushed the global embodied intelligence market to the eve of explosion. Various players and capital are rushing to seize this historic opportunity. UBTECH, Unitree Robotics, Galaxy General, and Zhiyuan Robotics have all announced plans to mass-produce thousands of humanoid robots this year.
Under the heated atmosphere, institutional hot money has been flooding into the embodied intelligence track this year. Industry investors indicate that since this year, the embodied intelligence field has frequently seen single financing rounds exceeding 500 million yuan, including Unitree, Galaxy General, Qianxun Intelligence, Star Era, Zhiju Dynamics, and Touchstone Intelligence.
An explosive new trend is gradually emerging. However, the gold-digging process is not overnight. How to fully integrate large models into embodied intelligence is currently a key bottleneck blocking AI's landing in the physical world.
Alibaba aims to be the "shovel seller" bridging this gap.
Previously, Wang Xingxing repeatedly stated that compared to hardware, embodied intelligence's greater challenge lies in large model shortcomings: "The biggest problem is that embodied intelligence AI isn't good enough."
Many large models performed brilliantly in virtual scenarios but struggled in the physical world. This is because traditional large models rely on massive internet text data for training, while embodied intelligence requires physical world interaction data, which is scarcer and harder to obtain.
Moreover, Gao Fei, head of embodied intelligence at Alibaba Cloud's Public Cloud Business Unit, points out that data volumes will steadily climb in coming years, with corresponding pressure on cleaning, storage, and training. "Relying solely on embodied intelligence companies' own strength, such pressure is hard to withstand."
Fortunately, Alibaba is prepared.
Currently, Alibaba's Qwen3-Max has over one trillion parameters, excelling in complex task decomposition, supporting robots in handling multi-step tasks like "unpack packages - sort items - arrange and store." Qwen3-VL has enhanced 3D spatial perception and dynamic object tracking, precisely identifying object positions and judging motion trajectories.
"Recently I've visited about 30 embodied intelligence companies and learned through deep conversations with their CEOs that most are using Qwen-VL models for post-training," Gao Fei says.
This common choice has solid technical reasons. Qwen-VL's capabilities in spatial perception, dynamic visual understanding, and path planning precisely match embodied intelligence's core "brain" requirements.
Additionally, Wang Xuwen, head of embodied intelligence solutions at Alibaba Cloud's Public Cloud Business Unit, notes that over the past few years, serving leading large model companies and automotive newcomers, Alibaba has "stepped on pitfalls step by step," accumulating substantial infrastructure and practical experience. Since embodied intelligence and autonomous driving share high architectural similarity, these experiences are transferable.
This full-chain support capability from model foundation to application layer gives Alibaba Cloud an almost unreplicable competitive advantage in the embodied intelligence track. Alibaba Cloud is helping clear obstacles for the entire industry on the path to mass-producing robots. When these capabilities transition from software simulation scenarios to the real world, embodied intelligence will truly enter explosive moments.
**Revaluing Alibaba: Full-Stack AI Loop**
Thus, Alibaba's ambitions become clear: Qwen serves as the universal "brain" responsible for task understanding and planning, while Alibaba-invested hardware companies serve as specialized "execution units" responsible for action implementation. Some even believe Alibaba has the potential to integrate its business with Pingtouge chip technology, potentially building an AppStore application ecosystem in the robotics field.
In this global game concerning the future, each player holds different cards. Alibaba has already cultivated a unique set of trump cards—a "full-stack AI" loop covering everything from computing power, cloud, models to application scenarios. This is not only its competitive capital but also the core of reshaping value and retelling its "greatness" narrative.
Looking at the market, we can see two distinctly different giant growth paths. Take SoftBank Group as an example—its founder Masayoshi Son is an excellent "hunter," rapidly assembling a global "information revolution" ecosystem through external acquisitions and investments.
In contrast, Alibaba's path is more like "cultivation," nurturing technological growth soil from within based on its massive business ecosystem. This model's advantage lies in natural connections between technology and scenarios. In the embodied intelligence era, these connections construct a unique flywheel effect.
This "full-stack AI" loop can be clearly deconstructed. Its foundation is Alibaba Cloud as a leading global cloud computing platform, providing massive, elastic computing resources for AI development and operation. Its brain is the Qwen large model, developed based on powerful computing power and tempered in multiple scenarios within Alibaba. Its body is robotic hardware and control algorithms being completed through internal team formation and external company investments.
The most core, hardest-to-replicate advantage of this loop is its unmatched testing ground—rich application scenarios. From Cainiao Network's millions of square meters of warehouses to Freshippo's automated stores to Taobao and Tmall's massive product processing, Alibaba possesses the world's largest, most complex real physical commercial scenarios.
These four layers jointly form the loop: real scenarios generate massive data; data inputs into Alibaba Cloud for computation and model training; optimized "Qwen" brains command robots; robots execute tasks in real scenarios and generate new data, further optimizing models in return.
Once this flywheel starts spinning, its iteration speed and efficiency will be remarkable. It enables AI models and robotic algorithms to rapidly evolve through real physical world feedback, far exceeding competitors limited to laboratory testing environments.
Therefore, betting on embodied intelligence is Alibaba's inevitable choice in attempting to become "great again" through AI narratives. Alibaba is not only a successful commercial company but also a hardcore technology company with foundational core technologies.
Industry investors point out that only full-stack approaches can make AI true infrastructure, enabling lower-threshold AI application proliferation across industries. Under full-stack AI's vast imaginative space, Alibaba's valuation logic receives additional momentum.
Recently, Goldman Sachs significantly raised Alibaba's 2026-28 fiscal year capital expenditure forecast to 460 billion yuan. Goldman Sachs believes AI capital expenditure conversion is reshaping Alibaba's growth expectations: "Alibaba's breakthrough progress in AI cloud computing capabilities and internationalization expansion potential provide new upward momentum for stock prices."
History's wheels roll forward relentlessly. Each computing platform transformation gives birth to new kings. Now we stand at the threshold of a new era jointly defined by AI and robots. Alibaba, the giant that once defined China's commercial era, is attempting to grasp the reins of this transformation by building a full-stack AI loop from cloud to ground.
This is undoubtedly a risky gamble, but for a giant yearning to be "great again," this may be the only correct path. The world is revaluing Alibaba, and the answer may be hidden in those "bodies" about to be endowed with intelligence.