Alibaba just told the world it wants more than 20 gigawatts of data centre capacity by 2032. CEO Eddie Wu dropped that number at the company’s Apsara Conference in Hangzhou, and it lands as one of the boldest infrastructure targets any Chinese tech company has put on paper. What Wu didn’t share is just as telling: no current capacity figure, no regional breakdown, nothing to measure the target against.
Here’s the part that actually matters. Wu admitted customer demand for AI services is already outpacing what Alibaba Cloud can supply. Supply-chain constraints, his words, are slowing how fast the company can add capacity. So the message from Hangzhou wasn’t really “we’re scaling.” It was closer to “we’re scaling as fast as the supply chain lets us, and that’s not fast enough yet.”
That tension runs through everything else Alibaba announced this week. The company is building its own AI chips through its T-Head unit, partly because Chinese firms need alternatives to Nvidia hardware under US export restrictions. The new Zhenwu V900 processor claims triple the performance of its predecessor from just four months ago, and Alibaba says it’s designing systems that support clusters of up to 500,000 accelerator cards. Impressive numbers on their own. But building your own chips because you can’t reliably buy someone else’s is also a workaround, not a victory lap.
Alibaba’s cloud revenue backs up the urgency. AI Cloud and Compute Services hit $7.1 billion last quarter, up 45% year over year, with AI-related products posting triple-digit growth for twelve straight quarters. That’s real demand, not a projection. The 380-billion-yuan investment commitment from February 2025 is already half spent, according to Reuters, months ahead of where anyone expected.
None of this guarantees Alibaba hits 20 gigawatts on schedule. Chip production, power availability, and global supply constraints don’t bend just because a company announces a target at a conference. What Wu’s comments actually reveal is a company racing to keep up with its own customers, building processors, networking, and storage in parallel because waiting on the usual supply chain isn’t an option anymore.
