U.S. Chip Curbs Escalate, Alibaba's AI Computing Expansion Hits a Wall

nashnova research
今天发布阅读约 8 分钟

A new U.S. Commerce Department draft rule aims to block Chinese cloud providers from leasing advanced chips through overseas data centers. Alibaba is a direct target — its market cap has halved from its peak to roughly $216 billion, and its access to AI compute is narrowing further.

01

What exactly does this new rule block?

Chinese cloud firms were already banned from buying Nvidia's most advanced chips directly. But a workaround remained: build data centers in Southeast Asia and lease high-end compute through offshore third parties.
The draft rule targets exactly that scenario. This means → not only is "buying" blocked, but "renting" is about to be shut down too.
Nvidia separately announced this week that it will stop shipping lower-tier Hopper data-center chips to mainland China, squeezing available compute options even further.
In plain terms = before, "the best chips were off-limits." Now, "the second-best won't be sold either — and you can't borrow them."
02

How far can Alibaba's overseas expansion still go?

Alibaba plans to build 20 gigawatts of data-center capacity by 2032, partly relying on overseas locations in Southeast Asia and beyond.
If the draft rule takes effect, that overseas expansion path faces direct policy-level uncertainty.
On the self-developed chip front, Alibaba's T-Head unit offers some buffer, but analysts say its chips still lag behind Nvidia's flagship products.
This reflects a core tension: in-house chips can narrow the gap, but they cannot close it in the near term.
03

What do analysts say — is the revenue hit significant?

JPMorgan analyst Alex Yao maintains an "overweight" rating, noting that Alibaba Cloud's overseas AI revenue accounts for less than 10% of its total cloud revenue — limiting the direct financial impact.
Nomura analyst Jialong Shi is more cautious on the broader control regime, but argues major cloud players are better positioned to absorb pressure than standalone AI labs.
He noted: "Standalone AI labs — perhaps excluding DeepSeek — may face greater challenges sourcing Nvidia GPUs to train next-generation models. Alibaba and ByteDance, with existing GPU stockpiles and stronger balance sheets, are in a relatively favorable position."
This means → the controls hit not "how much revenue Alibaba earns today" but "how powerful a model it can train tomorrow."
04

How much bad news is already priced in?

Alibaba trades at a forward P/E of roughly 10×, below its three-year average of nearly 13×, and in line with current valuations for Tencent and Baidu.
In plain terms = the market has already "priced in" a significant share of the bad news — but has not fully discounted the long-term impact of the controls.
Two verification points will determine the real depth of damage to Alibaba's AI strategy: whether Alibaba Cloud's overseas AI revenue share can grow, and whether its in-house chips can effectively substitute for Nvidia products.

市场有风险,内容仅供研究参考,不构成投资建议。