Alibaba unveils its proprietary AI accelerator chip (dubbed China's most powerful) and announces 10-trillion-parameter foundation model plan.

China's push to build advanced AI without dependence on American chips took its most concrete form this week through Alibaba. At its Apsara technology conference in Hangzhou, the company unveiled the Zhenwu V900, an AI accelerator it describes as the most powerful AI chip in China today, alongside a roadmap for future models far larger than anything it currently operates.

**The Chip**

Alibaba claims the Zhenwu V900 delivers three times the performance of its predecessor and carries 216 gigabytes of memory with high inter-chip bandwidth—the specifications essential for training large models. More strikingly, the company says a computing cluster built around the chip can link up to 500,000 of them for both training and inference. That scale is the threshold determining whether a country can train frontier models at all, and it targets the exact constraint US export controls were designed to impose. Mass production is scheduled for the first quarter of 2027.

**The Model Roadmap**

Alibaba outlined an ambitious trajectory for its Qwen models. The current flagship holds around 2.4 trillion parameters. Qwen 4 is in training, while Qwen 4.5 and Qwen 5 are projected to scale to between 5 and 10 trillion parameters—a substantial multiple of today's capability and comparable in scope to the largest models anywhere.

**Strategic Significance**

Announcing the chip and model roadmap together describes a self-contained stack: Chinese hardware training Chinese models at frontier scale. This outcome is precisely what export controls were designed to prevent—not to slow Chinese AI broadly, but to keep it dependent on Western chips and therefore controllable. A credible domestic accelerator that can scale to hundreds of thousands of units undermines that leverage and signals China's intent to circumvent the restrictions.

**Important Qualifiers**

These remain Alibaba's own claims announced at its own conference. "Most powerful in China" is a domestic comparison, not a claim to match Nvidia's latest offerings. The chip does not ship until 2027, leaving real-world performance and yield unproven. Parameter counts of 5 to 10 trillion are targets for unreleased models, not delivered capability—roadmaps announce intent, and intent is inexpensive to declare. The specifications also leave open a critical question: how efficiently 500,000 domestic chips can actually be made to work together, a challenge spanning software, networking, and silicon, where Western systems still retain advantages.

**What To Watch**

Whether the Zhenwu V900 ships on schedule in early 2027 and performs as claimed under independent testing. Whether Alibaba successfully trains a multi-trillion-parameter Qwen model on domestic hardware—the real proof of viability. And how Washington responds to a major Chinese company openly building the self-sufficient AI stack that export controls were intended to deny.