Alibaba has significantly escalated its push for domestic silicon independence with the unveiling of its next-generation artificial intelligence processor, the Zhenwu M890. Announced at an industry event in Chongqing, the new chip represents a major leap in computing capability for the e-commerce and cloud giant, arriving at a critical moment as Chinese technology firms face tightening restrictions on access to advanced foreign hardware.
According to CNBC, Alibaba stated that the Zhenwu M890 delivers three times the performance of its current flagship processor, the Zhenwu 810E. The new silicon boasts impressive specifications designed specifically for the heavy workloads of training and inferencing large language models, featuring 144 gigabytes of GPU memory and an interchip bandwidth of 800 gigabytes per second.
The announcement underscores Alibaba’s aggressive strategy to build a vertically integrated AI ecosystem, controlling everything from the underlying silicon to the cloud infrastructure and the foundation models themselves. This approach is increasingly necessary as U.S. export controls continue to choke off the supply of high-end Nvidia and AMD processors to Chinese data centers.
Scaling Domestic Compute
Alibaba is not merely designing chips for internal research; it is rapidly commercializing its silicon to capture the booming domestic market for AI computing power. The company revealed that it has already delivered 560,000 Zhenwu units to more than 400 customers across 20 different industries.
This massive scale of deployment indicates that Chinese enterprises are increasingly willing to adopt domestic hardware solutions, driven by both necessity and the improving performance of homegrown chips. The rollout of the M890 builds upon Alibaba’s earlier infrastructure milestones. In April, the company launched a 10,000-card AI computing cluster powered by Zhenwu AI chips, demonstrating its ability to network its proprietary silicon at a scale required for frontier model training.
Furthermore, Alibaba’s hardware ambitions extend beyond traditional data centers. The company’s chip design unit, T-Head, has also been innovating at the edge, recently releasing the XuanTie C950 RISC-V chip targeting agentic AI workloads. This dual-pronged approach, dominating massive cloud clusters with Zhenwu and pushing into edge devices with XuanTie, positions Alibaba as a comprehensive silicon provider in the AI era.
Teasing Qwen3.7-Max
Alongside the hardware announcement, Alibaba used the Chongqing event to tease the upcoming release of its next-generation large language model, Qwen3.7-Max. While specific performance benchmarks were not detailed, the “Max” designation in the Qwen family typically denotes the largest, most capable dense model designed to compete directly with global frontier models like GPT-5 and Claude.
The Qwen series has been a cornerstone of Alibaba’s AI strategy, achieving massive adoption globally. The company recently celebrated a major milestone when the Qwen family surpassed 1 billion downloads, capturing over 50% of the global open-source AI market. However, Alibaba has also shown a willingness to pivot its monetization strategy, recently shifting toward closed-source AI models to drive revenue. It remains to be seen whether Qwen3.7-Max will be released under an open-weights license or kept proprietary behind an API paywall.
The simultaneous announcement of the Zhenwu M890 and Qwen3.7-Max is not coincidental. It highlights the symbiotic relationship between hardware and software in modern AI development. By co-designing its chips and its models, Alibaba can optimize performance at the system level, potentially extracting more efficiency from its silicon than companies relying on off-the-shelf hardware.
The Broader Infrastructure Race
Alibaba’s hardware push is part of a broader national effort to secure China’s computing infrastructure. In early April, Alibaba and China Telecom announced the launch of a data center in southern China powered entirely by Alibaba’s proprietary chips. This collaboration between a private tech giant and a state-owned telecom operator exemplifies the coordinated push to build a resilient, domestically sourced computing network.
The demand for such infrastructure is surging. As China’s token economy explodes, with daily AI calls jumping from 100 billion to 140 trillion in 14 months, the need for cost-effective, high-performance inferencing chips has never been greater. The Zhenwu M890, with its 3x performance leap, is positioned to meet this demand head-on.
As the global semiconductor landscape fractures along geopolitical lines, Alibaba’s Zhenwu M890 serves as a critical proof point for China’s tech sector. It demonstrates that despite restricted access to the absolute bleeding edge of global semiconductor manufacturing, Chinese firms are capable of designing and deploying highly capable AI accelerators at scale, ensuring that the country’s AI ambitions are not derailed by hardware bottlenecks.