Woofun AI reports that Alibaba's shares tumbled 5% during trading in Hong Kong on June 25, sinking to HK$94.55 per share and marking a new low since February 2025. The market panic was not triggered by poor financial performance but by a letter sent by the American AI company Anthropic to Congress and the White House. In this correspondence, Anthropic accused Alibaba's Tongyi Qianwen team of utilizing nearly 25,000 fake accounts to interact with the Claude model over 28.8 million times, an action characterized as "large-scale model distillation." Anthropic labeled this incident "the largest known case of unauthorized model distillation to date."

Model distillation remains a standard technique within the AI sector where users compile output results from repeated queries into datasets for training new models. All major AI research institutions, including Anthropic itself, openly admit to systematically evaluating and comparing the capabilities of competing models. Previously, Anthropic had levied similar accusations against DeepSeek, Moonshot AI, and MiniMax. Equating normal model usage with "theft" appears to be an attempt by one company to frame its competitors unfairly, yet stock prices do not distinguish between technical merits and faults; they only respond to perceived risks. The immediate financial reaction underscores how quickly narrative shifts can impact valuation regardless of technical reality.

A more critical variable is the timing of these events relative to Alibaba's legal maneuvers. On June 23, Alibaba had just filed a lawsuit in a federal court in California, demanding that the US Department of Defense remove it from the "List of Chinese Military Enterprises." Two days later, Anthropic's accusation was widely reported by the media. The sharper the technological edge, the more intense the external pressures faced. Over the past six months, from internal conflicts to talent battles, from regulatory reviews to geopolitical issues, every advancement of Alibaba's AI efforts has been accompanied by factors unrelated to technology. For Alibaba's AI division, this is essentially a continuation of the sanctions it has been facing for two years.

Structurally, the geopolitical friction extends to high-profile partnerships. In February 2025, Apple and Alibaba announced a partnership to launch Apple Intelligence for iPhones in mainland China, powered by Alibaba's latest AI models. This seemed like a win-win arrangement: Apple needed a compliant local AI partner, and Alibaba needed a massive consumer base.

However, the cooperation did not proceed smoothly. Several jointly developed AI products had been submitted for review by the relevant authorities, but the process was stalled at the Cyberspace Administration of China. Sources familiar with the matter said that due to increasing geopolitical uncertainties between China and the US, Beijing was taking more time to review transactions or partnerships involving American companies, especially in key areas like AI. Ultimately, these transactions required approval from higher-level authorities, and as of June 2025, the approval process remained stalled. Apple sought Alibaba's cooperation in order to get the necessary approval, but the escalating trade tensions between China and the US led to more stringent reviews of their partnership. A simple technical collaboration ended up being hindered by tariff negotiations.

Notably, Alibaba's AI efforts faced direct challenges in the international market beyond regulatory delays. In September 2025, the American AI company Anthropic announced that it would stop providing its AI services to "Chinese-controlled enterprises." This was the first time an American AI company had made such a policy change, and it could affect Chinese companies such as ByteDance, Tencent, and Alibaba. Anthropic estimated that this policy would result in a loss of hundreds of millions of dollars in global revenue for these companies. In November, on the launch day of the Qianwen APP, foreign media cited a White House national security memo claiming that Alibaba provided technical support to Chinese military operations targeting targets within the United States. After lengthy attacks on the security of Chinese AI technologies using ambiguous language such as "possible" and "internally circulated", the report suddenly shifted focus, stating that there was no "factual verification" of these claims. This was almost a repeat of what happened to Huawei back in the past, using the guise of "national security" to carry out actual market sabotage. These geopolitical disruptions had nothing to do with Alibaba's technical capabilities but significantly affected the pace of its global expansion and its valuation in the capital market.

Woofun AI data shows that despite these external headwinds, Alibaba Cloud's AI-related products have seen triple-digit growth for eleven consecutive quarters, with revenue reaching 8.971 billion yuan in the fourth quarter of the 2026 fiscal year, accounting for over 30% of its external commercial revenue for the first time. Wu Yongming revealed at the earnings conference that the annual recurring revenue from AI models and application services would exceed 10 billion yuan in the June quarter. In Gartner's generative AI assessment, Alibaba Cloud ranked in the leader category in all four dimensions. An IDC report showed that Alibaba Cloud held a 42.2% market share in the public cloud market for large-scale model training and deployment in China. The Qianwen series occupies an important position among open-source models globally, with Qwen 3 achieving 1,433 points in the LMSYS Chatbot Arena, ranking third in the world. From chips (Pingtouge) to cloud infrastructure (Alibaba Cloud), from model development (Tongyi Qianwen) to application scenarios (e-commerce, DingTalk, Gaode), Alibaba has built a complete end-to-end AI ecosystem. Jack Ma referred to this as the "end-to-end AI strategy": developing its own chips, infrastructure, advanced large-scale models, and application solutions, and taking control of and optimizing each aspect of this ecosystem. This approach seems flawless on paper, but the completeness of the technical framework does not guarantee smooth business progress.

If the pressure coming from across the Pacific Ocean can be considered an external factor, then internal organizational conflicts represent the biggest obstacle Alibaba has faced recently. On June 4, former AI product manager of DingTalk's Wukong division, Teng Yaxin, posted a 75,000-word resignation letter on Alibaba's internal network, detailing the entire process of DingTalk's flagship AI project ONE from its inception to its eventual dissolution. Her main accusations were that the team catered too closely to higher-ups, engaged in excessive internal competition, and let leadership decisions override product logic. Six days later, Alibaba's Partner Committee publicly criticized DingTalk's management style in an internal announcement. Subsequently, DingTalk CEO Chen Hang resigned, and Chen Yusen, who was born in 1992, took over. This sudden management change had some element of luck, but it also revealed deep-seated organizational problems within Alibaba's AI division: when a flagship product is tasked with simultaneously achieving multiple goals, such as reducing user burdens, upgrading products, reforming the organization, and generating commercial revenue, product logic often gives way to political considerations. During Chen Hang's tenure, DingTalk launched dozens of AI products, including AI note-taking, AI search, AI spreadsheet tools, ONE, and Agent OS, resulting in a proliferation of functions. The most critical criticism of ONE in Teng Yaxin's letter was that it was actually designed to serve high-net-worth managers rather than ordinary employees. This exposed a fundamental disagreement in the product's purpose. Such disagreements were amplified and politicized within the organization, ultimately leading to the CEO's resignation.

Within three months, Alibaba made three adjustments to its AI organizational structure. In March 2026, it established the Alibaba Token Hub business unit; in April, it upgraded its technical committee; and in June, it merged the Tongyi Large Model division and the Future Life Lab to form the Token Foundry business unit, which was placed under Wu Yongming's direct supervision. Each of these adjustments aimed to address the same issue: how to ensure that Alibaba's AI initiatives could gain sufficient decision-making speed and resource priority within such a large organization.

However, frequent structural changes also drained the organization's execution ability and undermined the stability of its talent pool. At the beginning of March, the Tongyi Lab planned to split the Qwen team from a vertically integrated structure into horizontally organized units, with pre-training, post-training, text processing, and multimodal tasks each managed independently. As a result, Lin Junyang, who had advocated for a "small-team, large-loop" end-to-end approach, left the team. According to reports, after the successful launch of HappyHorse, the team members received numerous calls from headhunters. Companies such as ByteDance, Tencent, and several AI startups approached team members, with Tencent offering salaries for AI professionals that were roughly half higher than the market average. Zhou Jingren, a key figure in the Tongyi Large Model project, was promoted to chief scientist and no longer oversaw specific product lines. In the tech community, there was disagreement over whether this promotion was a form of recognition or a demotion.

The core dilemma facing Alibaba's AI division is that although its technical capabilities are constantly improving, the path to transforming these technologies into commercial success is frequently disrupted by non-technical factors. In 2026, the global AI commercialization landscape began to show clear divisions. Programming and video applications became the two main areas of focus. Cursor's annual revenue exceeded $2 billion, Anthropic's Claude Code accounted for 54% of the market share in programming scenarios, and ByteDance's Seedance 2.0 generated over 1 billion yuan in revenue for Volcano Engine in the video industry. Alibaba's own Tongyi projects also focused on programming and video, but few of its commercial products have gained widespread market recognition and adoption, aside from receiving praise from technical enthusiasts in the open-source community. in 2025, Volcano Engine held a 49.5% market share in token usage in China's enterprise-level MaaS market, while Alibaba Cloud had a 28% share, a gap of 20 percentage points. This difference is not due to technical capabilities but rather to differences in productization, scenario adaptation, and commercialization skills, and these aspects are particularly vulnerable to disruption caused by internal conflicts, strategic indecision, and external political pressures.

In June 2026, Alibaba released its first native language world model, Qwen-AgentWorld. In the same month, all three embodied intelligent models in the Qwen-Robot Suite were made open-source. Judging by the pace of technological development, Alibaba's AI efforts are indeed ongoing.

However, whether these technical advantages can be translated into market success depends on more complex factors: whether Alibaba can minimize the impact of these external distractions. After all, geopolitical tensions will not disappear just because of a company's technical strength, and internal organizational conflicts will not resolve simply because a CEO takes personal charge. Just a few days ago, the efforts invested by Jack Ma and his senior management team will still take more than a hundred days to bear fruit. The journey from planting seeds to reaping rewards in the AI field is also marked by long waits. But while the challenges in a rice field come from weather and pests, Alibaba's AI efforts often face variables that have nothing to do with technology or products themselves. In the world of agriculture, the principle is simple: you must plant when it's time to plant and endure when necessary. Whether Alibaba can overcome these external distractions will ultimately determine the success of its AI efforts. This marks a critical juncture where technical prowess must navigate a labyrinth of geopolitical and organizational constraints to achieve commercial viability.