Two cloud teams, one awkward board question
In Hangzhou and Beijing, two very different cloud organizations faced the same uncomfortable question from their boards: when does the AI story show up as revenue? For years the answer was a roadmap slide. In their latest earnings, both Baidu (百度) and Alibaba (阿里) Cloud found an answer they could actually put in the numbers — and it came from selling inference, not hype.
The shared pattern is simple. Enterprises that once dipped a toe into large models (大模型) are now renting the compute, storage and model-serving layers to run them in production. That demand is the quiet engine behind both companies' cloud reacceleration.
Baidu: the AI cloud that finally compounded
Baidu reported its fiscal 2025 results in February 2026, and the "AI-powered business" line — a grouping it created to track exactly this shift — told the story clearly.
- Baidu Core's AI-powered Business reached RMB 40.0 billion (≈ US$5.63B / HK$43.6B) for the full year, up 48% year on year.
- Within it, AI Cloud Infra revenue was RMB 19.8 billion (≈ US$2.79B / HK$21.6B), up 34%.
- In the fourth quarter alone, AI Cloud Infra hit RMB 5.8 billion (≈ US$817M / HK$6.32B), and subscription revenue from AI accelerator infrastructure jumped 143% year on year.
- AI-native marketing services — digital humans and agents — more than tripled to RMB 9.8 billion (≈ US$1.38B / HK$10.7B), up 301%.
- The AI-powered business was 43% of Baidu's general business in Q4, up from 39% for the full year.
Baidu also pushed its own stack deeper: it released ERNIE 5.0 (文心 5.0), a native omni-modal model with 2.4 trillion parameters, in January 2026, and filed to spin off its Kunlunxin (昆仑芯) AI-chip unit in Hong Kong.
Alibaba: triple-digit AI growth, ten quarters running
Alibaba's cloud story is bigger in absolute terms and more consistent in cadence. In its December-quarter 2025 results (reported March 2026), the Cloud Intelligence Group posted:
- Revenue of RMB 43.28 billion (≈ US$6.10B / HK$47.2B), up 36% year on year — the fastest cloud growth in years.
- AI-related product revenue delivered triple-digit year-on-year growth for the tenth consecutive quarter.
- In the prior (September) quarter, cloud revenue was RMB 39.82 billion (≈ US$5.61B / HK$43.4B), up 34%, with AI products up triple digits for the ninth straight quarter.
The demand is broad-based. According to Omdia's 1H25 China AI-cloud market report, Alibaba Cloud holds the #1 position with a 35.8% share — larger than the next three vendors combined. Its consumer Qwen (千问) app surpassed 300 million monthly active users by February 2026, and the Qwen model family crossed 1 billion cumulative downloads on Hugging Face as of January 2026.
What "AI lifting cloud" really means
The mechanism is the same at both companies, even if the scale differs. Generative AI turns a one-time model download into a continuous bill:
- Training and fine-tuning consume GPU clusters by the hour.
- Inference — serving a model to millions of users — is a recurring, sticky workload.
- Model-as-a-Service (MaaS) platforms let enterprises call a model without owning the stack, which is exactly the layer both Baidu and Alibaba are expanding.
Alibaba explicitly called its MaaS platform a "new engine" for cloud growth, and Baidu's 143% surge in accelerator-infrastructure subscriptions is the same phenomenon seen from the supply side. When a model goes from a demo to a product, the cloud bill does not go away — it becomes the product.
The catch investors should not ignore
Neither company is printing money on this yet. Alibaba's December-quarter operating income fell 74% year on year, driven by heavy investment in quick commerce and AI infrastructure; free cash flow dropped 71%. Baidu's total revenue was roughly flat (down slightly to RMB 129.1 billion / ≈ US$18.2B / HK$140.7B) as its older search-ad business eroded even as AI rose.
Both are also spending aggressively to keep up: Alibaba said its capital expenditure on AI and cloud reached about RMB 120 billion (≈ US$16.9B / HK$131B) across the four quarters through September 2025. The AI cloud boom is real, but it is being bought with margin and cash flow today in exchange for a larger position tomorrow.
Honest limitations
- Figures are drawn from each company's own earnings releases and SEC/HKEX filings (Baidu IR; Alibaba Group / SEC EX-99.1). They are management-reported and denominated in RMB; USD/HKD conversions use approximate rates (÷7.1, ×1.09) and are flagged as such.
- "AI-related product revenue" is a company-defined category, not a standardized accounting line; Baidu's "AI-powered Business" and Alibaba's "AI-related" segments are not directly comparable to each other.
- Market-share data (Omdia 35.8%) is a third-party estimate for the first half of 2025 and may not match other research firms' definitions of "AI cloud."
- Profitability context (margin and FCF declines) is from the same filings but reflects mixed business lines, not cloud alone, so cloud margin cannot be isolated from these releases.
- No figure is invented or rounded into a hard claim; where a number is a growth rate rather than an absolute, it is presented as a rate.
What readers can do now
- If you buy cloud for a Chinese operation, benchmark both ERNIE and Qwen serving costs against your current provider — the 34–36% cloud growth at both firms is partly a price-and-capability race you can exploit by testing before committing.
- Track the "AI Cloud Infra" and "AI-related product" lines specifically in each quarterly release; they are the cleanest signal of whether enterprises are moving from piloting models to running them in production.
- Watch capital expenditure, not just revenue. Both firms are sacrificing near-term cash flow for AI capacity — if their capex stays ahead of AI-cloud revenue for too many quarters, the "AI lifts cloud" story gets harder to finance.