Alibaba’s Zhenwu AI Chips Power China’s AI Data Center Frenzy—But Can It Avoid the Overcapacity Trap?

The market is fixated on a single, massive trend: China's aggressive push to build its own AI infrastructure. This isn't theoretical; it's a viral news cycle dominated by announcements of colossal data center projects and record-breaking compute clusters. The catalyst is clear: Beijing's drive for technological self-reliance is translating into a frenzy of domestic investment, and Alibaba is positioned as a key beneficiary.

The most concrete signal is Alibaba's own move. The company has signed an agreement with the Jinshan District government in Shanghai for a new data center, part of its "Aspara Cloud Intelligent East China Computing Center." This facility will house Alibaba's custom Zhenwu AI chips, a direct play on the national strategy to build a full-stack, independent computing foundation. This isn't a side project. Alibaba is already scaling its AI capex, with plans to invest $69.05bn (480 billion yuan) over the next three years, a significant jump from earlier promises.

This fits a broader, intense demand picture. Just weeks ago, China unveiled its first 10,000-card AI cluster built with Huawei Ascend chips. This project, now reaching 14,000 petaflops, shows local demand is no longer a future promise but a present reality, with a 92% booking rate indicating a severe shortage of compute power. The context is a race to close a hardware gap, using scale and software to compensate for chip efficiency that lags behind Western counterparts.

For investors, the search interest and news cycle are the real indicators. The dominant theme is China's massive, state-backed buildout of AI data centers. Alibaba's role here is central. It's not just a customer; it's a builder, a chip designer, and a cloud provider, all rolled into one. In a market where Alibaba Cloud holds more than a third of the local AI cloud market, its ability to deploy custom hardware at scale gives it a unique advantage. The company is the main character in this domestic infrastructure story, translating national ambition into concrete projects and, potentially, revenue.

The Tech Stack: Chips, Software, and the Efficiency Gap

Alibaba's push is a full-stack play, but it operates on a different efficiency curve than the global leader. The company is targeting a $100 billion combined cloud and AI external revenue within five years, a goal it believes is within reach given its recent momentum. That momentum is real: the Cloud Intelligence Group grew 35 percent year-on-year last quarter, hitting $6.19 billion. This isn't just about selling cloud storage; it's about selling AI capability, with AI product revenue showing triple-digit year-over-year growth for the 10th consecutive quarter.

To build that capability, Alibaba is designing its own chips. For training and inference, it's rolling out its Zhenwu AI chips, with over 470,000 shipped as of February 2026. For a newer frontier-agentic AI that can carry out multi-step tasks-it has unveiled the XuanTie C950-5 CPU. This chip is built on the open-source RISC-V architecture, a deliberate move to avoid licensing fees and customize for specific workloads. Alibaba claims it can achieve over 30% improvement in performance thanks to that flexibility, and it's being positioned to power platforms like its new Wukong agentic system.

Yet the core challenge remains the same as China's broader AI race: hardware efficiency. While Alibaba's chips are scaling, they still lag behind foreign counterparts in performance. This mirrors the situation with Huawei's Ascend chips, which run at about 60% of an Nvidia H100. The gap is real. For now, the strategy is to compensate with scale, system design, and tight integration between custom hardware and software like the Qwen model. Alibaba is betting that its full-stack control-chips, cloud, and AI models-can deliver better cost-effectiveness, even if raw speed per chip is lower. It's a race of systems over single components.

Market Attention vs. Headline Risk: The Bubble Watch

The market's attention is laser-focused on China's AI buildout, and the demand signal is extreme. A new AI cluster built with Huawei chips is already 92% booked, showing local labs and universities are desperate for compute. This isn't a future projection; it's a present-day shortage. Yet, amid this frenzy, a major warning has emerged. Alibaba's chairman, Joe Tsai, has stepped into the spotlight to caution against the rush. At the HSBC Global Investment Summit, he said he starts to see the beginning of some kind of bubble in speculative data center construction. His concern is specific: projects are being funded without customer contracts, and capital is flowing in ahead of real demand.

The primary risk to the bullish thesis is now clear: overcapacity. If the current wave of investment-driven by both state ambition and private capital-outpaces the actual need for AI compute, it could trigger a painful correction. This isn't theoretical. Just last month, Microsoft reportedly canceled data center leases worth hundreds of megawatts, a move analysts saw as a potential early sign of oversupply. The bubble watch is on.

For Alibaba, its main character status hinges on its ability to navigate this risk. The company is not just a builder; it's a chip designer and a cloud provider. Its success depends on executing its full-stack roadmap. The recent shift in the chip market is a key test. Huawei's older flagship chip struggled to gain traction with big tech firms, but its new 950PR chip is more compatible with Nvidia's CUDA software system, making it far more appealing. Sources say giants like ByteDance and Alibaba plan to place orders. This move from proprietary software to industry-standard compatibility is critical. It shows the market is demanding practicality over pure nationalism. Alibaba's ability to secure orders for its own Zhenwu chips, or to effectively integrate and sell Huawei's more compatible 950PRs, will determine if it can convert the current hype into sustainable revenue. In a market where attention is high but bubble fears are rising, execution is the ultimate catalyst.