Alibaba Is Renting the Chips That Built a Rival-And the Market Still Cheers

Alibaba's AI strategy is creating a rival while strengthening its cloud role

Alibaba is increasingly looking like China's AI compute layer, but Moonshot may be the first major winner. Reports say Alibaba has a computing agreement with Moonshot for about 20,000 Nvidia chips, and Alibaba did not deny that broader claim even as it pushed back on the idea that H200 chips were specifically provided.

That tension is the core of the story. Alibaba is monetizing scarce capacity, yet Kimi uses Alibaba infrastructure, and Moonshot's Kimi has outperformed Qwen on some important metrics. In practical terms, Alibaba may be building the roads while Moonshot drives the faster car.

The picks-and-shovels upside

Bulls see a straightforward infrastructure trade. If AI compute stays tight, Alibaba can keep renting capacity and deepen its role as a cloud supplier. That demand appears to be helping the business now, with cloud revenue growing 38% as AI demand lifted the segment.

Why the alignment is still shaky

The bear case is that Alibaba may be helping a rival become stronger. The company is one of Moonshot's biggest backers, but the two firms also compete in China's AI market. Moonshot is not a failed experiment either: Kimi K3 went toe-to-toe with leading models, even if it still trails the very top U.S. systems overall.

The market's reaction captured that tension. After adjusted EBITA fell 84%, Alibaba shares fell as much as 4% in pre-market trading before rallying 7.5% as investors focused more on AI demand than near-term profit pressure.

Alibaba's cloud bet is about more than raw compute

The market is not just paying for one partnership. It is betting on a broader platform shift.

Alibaba's cloud business is no longer selling compute like a utility. It is selling an AI stack that layers infrastructure, model hosting, and agent tooling together. Management said AI-related products accounted for 30% of cloud external revenue growth, while external cloud revenue rose 40%. That suggests AI demand is showing up in actual customer spending, not just headline benchmarks.

Why the model could work

The logic is familiar: once enterprises build workflows on a cloud platform, switching can become costly. Alibaba is leaning into that dynamic. At WAIC 2026, it launched an Agent-Native Cloud suite to orchestrate multi-agent workflows, along with developer tools and the open-source T-Head SAIL software stack. The idea is simple: Alibaba does not have to win every model leaderboard if customers deploy, tune, and monitor agents on its platform.

That platform angle also explains the market's willingness to look past weak near-term profitability. Alibaba is trying to turn AI demand into a broader stack sale, not just a few blocks of GPU hours.

The stress test from Moonshot

Moonshot also shows how hard platform loyalty can be to secure. Alibaba may control the infrastructure, but Kimi uses Alibaba infrastructure, and Moonshot still delivered meaningful performance gains despite China's hardware and compute capacity constraints. That cuts both ways.

It validates Alibaba as a key layer in China's AI stack, but it also shows that a model team can use that layer and still keep the more valuable customer relationship. If top U.S. models remain ahead overall, bears will argue that some AI workloads in China still are not mature enough to support durable, high-margin contracts.

Management has signaled that the payoff window is the next three to five years, with returns expected to become much clearer over that period. If that timeline holds, Alibaba is being valued less as a chip renter and more as a foundational AI platform. If it slips, the current optimism will be harder to defend.