Alibaba's decision to block Anthropic's Claude Code for employees has intensified the competition among Chinese tech giants to develop homegrown AI coding models, with market volatility and investor stakes raising the pressure on rivals
Two companies now dominate the global race to build the most advanced AI coding models: Anthropic, with its Claude Fable 5, and Alibaba, which has just previewed its Qwen 3.8 Max system. The rivalry is not just technical-it's increasingly financial and geopolitical, with both sides maneuvering for advantage as U.S.-China tech tensions persist. According to Bloomberg, Alibaba's shares jumped as much as 5.4% in Hong Kong on July 20 after the company claimed its new model is second only to Fable 5, a benchmark that has become the standard for Chinese AI labs.
The timing of Alibaba's announcement is notable. Just three days earlier, Moonshot AI's Kimi K3 made a similar claim, briefly unsettling global tech stocks. But Alibaba's move comes with added complexity: the company recently banned its employees from using Anthropic's Claude Code for work, citing concerns after discovering code that could detect Chinese users. This ban, effective July 10, followed Anthropic's testimony to the U.S. Senate Banking Committee that operators linked to Alibaba's Qwen lab had run about 25,000 fraudulent accounts to extract Claude's outputs using a process called distillation.
Internal Competition and Market Impact
Alibaba's internal ban forced staff to switch to its own coding platform, Qoder, which now hosts the Qwen 3.8 Max preview. The company is effectively racing to replace the very tool it just prohibited, while also trying to match or surpass Anthropic's performance. This dynamic is not unique to Alibaba. Zhipu AI, another Chinese lab, made a similar claim in June, stating its GLM 5.2 model nearly matched Opus 4.8 on a key benchmark at a lower cost. Moonshot AI's Kimi K3, launched July 17 with 2.8 trillion parameters, also positioned itself as a direct challenger to Anthropic, and is reportedly planning an IPO within six months.
Despite these bold claims, none of the Chinese labs have published independent benchmark results for their models. Alibaba has yet to release a model card or detailed performance table for Qwen 3.8 Max, meaning its assertion of trailing only Fable 5 is based solely on internal testing. This lack of transparency is a recurring theme among Chinese AI challengers in 2026.
Open Weights and Access Risks
One of the most significant shifts in this AI arms race is the growing emphasis on open weight models-AI systems whose underlying code and parameters can be downloaded and run on private servers. Alibaba has promised to release Qwen 3.8 as an open weight model, though no date has been set. This approach is gaining traction after the U.S. briefly restricted foreign access to Fable 5 and its sibling model, Mythos, in June, only partially restoring access in July. For Chinese labs, offering open weights is a way to sidestep export controls and ensure continuity for users, regardless of international policy shifts.
Alibaba's financial interests are also intertwined with its competitors. The company holds a roughly 36% stake in Moonshot AI, giving it exposure to both sides of the contest to become the default open alternative to Anthropic. As a result, Alibaba's stock gains reflect not just a victory over outside rivals, but also the performance of its own portfolio companies. This dynamic is reminiscent of how cross-holdings can blur the lines between competition and collaboration in China's tech sector, much like the complex relationships seen in other high-profile tech rivalries, such as those discussed in analysis of Tesla and SpaceX's merger speculation.
Volatility Among AI Stocks
The market reaction to these developments has been swift and uneven. While Alibaba's shares rallied, other Chinese AI firms saw steep declines. MiniMax dropped 6% on the same day Alibaba surged, and Zhipu AI, which trades as Z.AI, fell as much as 15% after a 28% plunge in the previous session. According to Seeking Alpha, analysts at Goldman Sachs have identified a trend of large-parameter, high-end coding models competing head-to-head, with Alibaba named a top pick among cloud developers due to rising spending on AI infrastructure. The firm argues that Alibaba stands to benefit from increased demand for AI, regardless of which model ultimately leads the benchmark race.
For investors, the stakes are not just about which model is technically superior, but also about which company can guarantee access and continuity. The brief U.S. ban on Fable 5 and Mythos highlighted the risk that even top-performing models can become inaccessible due to policy changes. This has made the ability to download and run models locally-a feature of open weight systems-a key selling point for enterprise customers and investors alike.
Access Versus Capability
The contest between Alibaba and Anthropic is no longer just about raw performance. Access, reliability, and independence from foreign policy decisions are now central to how investors and customers value AI providers. Every new claim of trailing only Fable 5 generates a burst of media attention and a short-term stock boost, but Anthropic's leadership position remains unchallenged for now. Meanwhile, surging domestic demand is straining infrastructure, as seen when Moonshot AI had to halt new user sign-ups to manage server load.
Ultimately, the real question is which Chinese lab-and by extension, which of its backers-will become the default choice if access to Anthropic's models is restricted again. For now, Alibaba's stock is reflecting investor bets on that outcome, even as the competitive landscape remains unsettled.
According to Alibaba's most recent quarterly report, the company's cloud division generated $4.2 billion in revenue for the quarter ended March 31, 2026, up 12% year-over-year. This growth was driven in part by increased demand for AI infrastructure and services, as Chinese enterprises accelerate adoption of large language models and coding assistants. The company's investment in Moonshot AI and other startups has also contributed to its exposure in the rapidly evolving AI sector.
Open weight AI models represent a fundamental shift in how companies and governments approach technology risk. By allowing organizations to run models on their own servers, open weights reduce dependence on external providers and mitigate the risk of sudden access restrictions. This is especially relevant for multinational firms and regulated industries, where data sovereignty and operational continuity are critical. However, open models also raise new challenges around security, intellectual property, and compliance, as organizations must take greater responsibility for managing and updating the technology themselves. As the AI landscape evolves, the balance between access, capability, and control will remain a central issue for both investors and end users.
