Alibaba (BABA): Cloud Keeps Accelerating
38% cloud growth, margin expansion guided, quick commerce profitability pulled forward
Alibaba reported Q4 2026 earnings yesterday morning and the market finally got the kind of cloud commentary it has been waiting for.
The reported numbers were definitely a bit confusing/mixed. Revenue was fine, profitability was pressured, and quick commerce is still weighing on margins. None of that is new though. That has been the setup for the last few quarters.
What was different this quarter, or at least much clearer, was the shape of the business itself. The restructuring is mostly behind them, the divestitures are fading out, and you’re really starting to see Alibaba as just two businesses at this point: E-commerce and Cloud/AI.
And on the cloud side specifically, this was the quarter where management got a lot more direct about the AI revenue trajectory and where margins are heading.
Cloud
Cloud revenue grew ~38% year-over-year, with external revenue up roughly ~40%. Cloud EBITA was up ~57% on a roughly ~9% margin. Wu guided to further acceleration from here.
AI-related products hit RMB 9 billion in the quarter, triple-digit growth for the eleventh straight quarter, now representing about ~30% of external cloud revenue.
“We expect that in about one year, AI-related product revenue will cross the 50% threshold, becoming the primary engine driving cloud business revenue growth. As a result, Cloud Intelligence Group’s external revenue growth is expected to continue accelerating beyond its current 40% rate over the coming quarters.” — Eddie Wu
So the fastest-growing piece of cloud is about to become the majority of it. The math on where overall growth goes from there isn’t complicated.
Management also disclosed Model and Application Services ARR for the first time. RMB 8 billion-plus currently, guided to over RMB 10 billion by the June quarter and RMB 30 billion by year-end. And platform usage is up more than 10x since November. This finally gives the sell-side something concrete to model that they didn’t have before, and I think it changes how the street values the AI piece going forward.
On profitability, Wu was pretty explicit that margins would improve over the coming quarters.
The way they framed it, there are a few things working in the right direction: MaaS is becoming a bigger part of the mix and carries better margins than raw IaaS. T-Head custom silicon is scaling, with over 60% of capacity now serving external customers. And Alibaba appears to have real pricing power because compute is still scarce. New server replacement costs are reportedly running around 2x year-over-year, which tells you how tight the market still is.
Wu on what they’re seeing on the inference side:
“Every quarter we are seeing new results in terms of optimization in reasoning and inferencing, with continuous incremental effects in terms of the token capacity of a single server and a single card.”
And on pricing:
“At the same time as the capabilities of models continue to strengthen and prices of the models continue to increase in the next year or two, we see that this should be a process of continued price improvement.”
And then directly on margins:
“I think from this point of view the rapid growth in this business over the next few quarters will result in very positive impact on our overall gross profit margin.”
Inference getting more efficient quarter over quarter. Capacity running near full. Replacement costs doubling. Model prices going up. And management guiding to gross margin expansion in the next one to two quarters. That is a pretty good hand to be holding. Especially for a cloud business still accelerating at this size.
And as I expected, they also confirmed they’ll likely exceed the RMB 380 billion three-year capex plan they announced last year. No new ceiling given though.
Management expects roughly 10x compute demand growth by 2033 versus 2022, which is an enormous number, but when you consider that Alibaba holds about 40% of Chinese cloud market share (larger than the next three competitors combined) with a full vertical stack running from custom silicon all the way up to proprietary models, the spend starts to make more sense.
