Alibaba and Tencent Signal the Same: Capex to Decline After 2028 as Domestic Computing Power Takes Over

nashnova research
今天发布阅读约 14 分钟

Alibaba and Tencent gave near-identical guidance on their latest earnings calls: AI infrastructure spending peaks between 2027 and mid-2028, then declines, with domestically designed chips replacing imports as the core engine of cloud revenue growth.

01

When does the spending peak hit?

Alibaba CFO Toby Xu disclosed quarterly capex of RMB 50.5 billion, down from RMB 67.7 billion the prior quarter. The highest intensity falls in 2027 and H1 2028.
Tencent CFO James Mitchell reported Q3 operating capex payments of RMB 63.6 billion, up 257% year-on-year. Management expects cash capex to fluctuate around this peak level through 2026–2027, then decline.
This means → Both companies drew nearly the same curve: one to two more years of heavy spending, then a pullback. For the entire AI compute supply chain, H2 2028 is the first verifiable inflection point.
02

How did domestic chips go from backup plan to main bet?

Alibaba CEO Eddie Wu said supernode instances built on the in-house Zhenwu M890 chip have reached commercial scale. The next-generation V900 enters mass production in Q1 2027 with 3× the compute performance; a further J900 is planned for 2028.
Xu stated explicitly: Alibaba's AI infrastructure roadmap through 2030 has always been built on the T-Head domestic chip supply chain. Domestic compute's share will keep rising and "become the primary engine supporting cloud revenue growth."
Tencent management was even more direct: inference workloads — the computations a model runs when answering queries — already run efficiently on a surprisingly limited set of domestic chips. Over the past 12 months, inference productivity gains have let domestic chips handle far more tokens than anyone expected a year ago.
In plain terms = Domestic chips are no longer the fallback if imports get cut off. They are the primary roadmap both companies are actively betting on. Export-control impact was characterized as "short-term."
03

How fast can cloud revenue grow from here?

Alibaba guided next fiscal year (FY2028) cloud revenue growth at 30%–45%, below the market's prior expectation of 50%+. But once domestic compute scales through calendar 2028, management expects a re-acceleration to above 50%.
Alibaba's AI-related product revenue hit RMB 17.9 billion this quarter — the 13th consecutive quarter of triple-digit growth — pushing annualized AI revenue past US$10 billion. The 2030 target stands: US$100 billion in external cloud revenue, margins above 20%.
Tencent cloud services revenue growth accelerated to 35% year-on-year, up from 21% in Q2, driven by WorkBuddy, model-as-a-service — selling large models as callable cloud products — and GPU leasing.
This means → Both growth curves have a dip-then-reaccelerate shape in the middle. The dip comes while domestic compute is still ramping; the reacceleration arrives when the ramp is done.
04

How do in-house chips change the margin structure?

Alibaba management said in-house chips plus scale advantages create a procurement edge: "compared to running cloud and AI businesses on externally sourced chips, margins will see a positive uplift." Alibaba Cloud's adjusted EBITDA margin already expanded to 14.9% this quarter.
Alibaba also flagged a pricing-logic shift: in the AI era, Alibaba Cloud is no longer just a supply-chain vendor but a co-development partner with customers. That product differentiation makes direct price comparison difficult — a structural factor supporting margins long-term.
Tencent management said compute efficiency has roughly doubled year-on-year: the same facilities can produce more tokens over time. President Martin Lau added that by 2028, "even if you need to buy the same thing, you should expect the price to drop significantly."
In plain terms = In-house chips are cheaper, externally sourced chips are also falling in price, and the same hardware handles more workload — three lines pulling costs down simultaneously. Cloud gross-margin headroom is structurally opening up.
05

What exactly does the 2028 checkpoint verify?

Both companies' logic converges on a single question: can domestic compute scale on schedule through 2028?
If the ramp delivers → the capex inflection holds, free cash flow turns positive (Alibaba targets H2 2028), and cloud revenue re-enters the 50%+ growth lane.
If the ramp slips → peak spending drags on longer, cash flow stays under pressure, and the growth dip deepens.
This reflects a clear investment framework: the success or failure of both companies' AI narratives now hinges not on revenue growth rates, but on the domestic chip capacity ramp timeline.

市场有风险,内容仅供研究参考,不构成投资建议。