Alibaba's AI Cloud Profit Margin Rises to 12% as Proprietary Chips Serve Over 650 Clients
Nashnova编辑部
Alibaba Cloud's AI unit posted a 12% adjusted EBITDA margin in Q1, with EBITDA up 133% year-on-year and AI product revenue growing at triple digits for a 12th straight quarter; its in-house chip now serves over 650 external clients, signaling the shift from spending to profiting.
What does a 12% margin actually tell us?
The AI cloud and computing segment's adjusted EBITDA margin rose to 12%, with EBITDA up 133% year-on-year.
This means → AI spending is crossing over from a "burn cash, grab scale" phase into one where profits are starting to flow back.
CEO Eddie Wu framed this quarter as "accelerating growth with rising margins" — last quarter's label was merely "crossing the inflection point."
How fast is AI revenue growing?
Alibaba Cloud's external commercial revenue grew 45% year-on-year — the fastest pace in 22 quarters.
AI-related product revenue hit RMB 12.376 billion, marking a 12th consecutive quarter of triple-digit year-on-year growth.
AI product annualized recurring revenue (ARR) topped RMB 49.5 billion, accounting for 35% of external cloud revenue; management guides next quarter's ARR to reach US$10 billion (roughly RMB 73 billion).
In plain terms = for every three dollars of cloud revenue, more than one now comes from AI — and that share is still expanding fast.
Where does the in-house chip stand?
T-Head's Zhenwu M890 — Alibaba's self-designed AI inference processor — is now commercially deployed across 20+ industries and more than 650 external clients, spanning autonomous driving, finance, and internet services.
Over 500,000 units of the previous-generation chip have been produced and shipped; the latest generation went live on Alibaba Cloud in August as Supernodes — clusters of chips pooled into a single massive compute unit.
Zhenwu M890-based Supernodes can run inference for models exceeding 2 trillion parameters; Kimi K3 and Qwen 3.8 Max are already running on them.
This means → Alibaba's in-house chips are no longer just an internal substitute — they are selling compute externally, entering a commercial scale-up phase.
Where has the RMB 380 billion gone?
By end of Q1, roughly RMB 190 billion — exactly half — of the three-year RMB 380 billion investment plan announced in February 2025 had been deployed.
CFO Toby Xu's logic: AI monetization rests on compute infrastructure, and "capex must come first before business growth can follow."
In plain terms = spend on data centers and chips upfront, then client orders catch up — a "spend first, earn later" business.
How long until the investment pays back?
Management estimates AI compute assets can recoup costs in roughly three years.
Alibaba Cloud has compressed large-scale AI data-center delivery to 100 days and expects modular data-center capacity efficiency to more than double in 2026.
This reflects Alibaba's push to shorten the payback window by building faster and cheaper — but the market's real test remains: the pace of Zhenwu M890 Supernode ramp, and whether the AI cloud margin keeps climbing in coming quarters.
Content is for reference only, not financial advice.