Author | Hu Min, Editor | Liu Wei
After reviewing Alibaba's latest earnings report released last night, there is a fact one cannot help but accept, even if reluctantly: while the market's spotlight is on the model and application layers, in the short term, the most robust AI commercialization remains in cloud and chips. Moreover, the more intensively the upper-layer models are utilized, the more profitable the underlying "pick-and-shovel" providers become.
The latest earnings report shows that in the first quarter of fiscal year 2027 (the second quarter of calendar year 2026), Alibaba Cloud's revenue reached 48.737 billion RMB, achieving a year-over-year growth rate of 45%.
More importantly, there is the revenue guidance for the upcoming quarters. During the earnings call, Wu Yongming explicitly stated: Given the strong feedback on customer demand, they anticipate that Alibaba Cloud's revenue growth rate will continue to accelerate in the coming quarters.
This essentially reassures investors: there is no need to worry that this quarter's growth is merely a flash in the pan; the momentum will remain strong in the upcoming quarters.
01. Where Does Alibaba Cloud's Revenue Growth Come From?
Delving deeper, the driving forces behind Alibaba Cloud's growth can perhaps be divided into two aspects:
First, the training and inference of upper-layer models and applications drive massive consumption of AI computing power resources.
At the end of last year, Leiphone mentioned in the article "Cloud Giants 2025 Review: Breaking Through the Vicious Cycle of 'Growth Deceleration'" that in 2025, cloud giants collectively saw a surge in performance driven by GPUs. Preliminary research at the time also showed that Alibaba Cloud's AI product revenue grew rapidly, making it the fastest-growing product among all its offerings.
Moving into this year, the demand for AI computing power has only increased, making the market even tighter for computing resources. A computing power practitioner told Leiphone that, compounded by U.S. sanctions and tightening domestic regulations, the overall supply gap for computing power has continued to widen in the first half of this year.
During the earnings call, Wu Yongming echoed this sentiment, stating, "Before 2030, it is fundamentally difficult to see the tight supply of AI computing power being resolved."
In this state of supply falling short of demand for computing power, Alibaba also addressed the highly concerning issue of capital expenditures. The subtext of Wu Yongming's remarks is: there is little need to worry about Alibaba's capital spending being too high at present; before 2030, the return on investment for AI computing power possesses high certainty.
In February last year, Alibaba took the lead in the industry by proposing a three-year investment plan of 380 billion RMB. This quarter's capital expenditure reached 67.1 billion RMB. As of the second quarter of 2026, Alibaba's CAPEX has reached 190 billion RMB, with the progress basically in line with expectations.
"For the remaining capital investment this year, Alibaba will continue to maintain a more aggressive stance."
Second, intelligent computing drives the consumption of general computing. With the emergence of AI, all customers are learning, using, and purchasing AI. As the application of AI deepens progressively, it will drive customers' cloud usage demands.
In South China, there is a vast traditional manufacturing sector. A cloud sales representative told us about a client that had been doing PCB (Printed Circuit Board) contract manufacturing in Dongguan for twenty years. Originally, they had almost no cloud usage needs, relying only on a few local servers. Later, with the advent of AI, they attempted to train an AI quality inspection model. During the execution process, they gradually purchased a series of products such as cloud storage and cloud databases.
Some analytical reports can also verify this. According to Gartner data, Alibaba Cloud's share in China's IaaS market rose from 30.1% in 2024 to 32.8% in 2025, while Huawei Cloud and Tencent Cloud both experienced varying degrees of decline during the same period.
02. Profit, Another Unexpected Data Point
Of course, besides revenue, another highly unexpected figure in this quarter's earnings report is Alibaba Cloud's profit. This quarter, Alibaba Cloud's adjusted EBITA grew by 133% year-over-year, and its profit margin rose to 11.6%.
This is a figure that greatly boosts industry confidence.
Previously, the low profit margins of domestic cloud vendors have been widely criticized. For a long time, the market has barely factored Alibaba Cloud into its valuation.
This quarter, Alibaba Cloud's profit margin rose to 11.6%, which has a boosting effect on the valuation of the entire domestic cloud sector.
Delving deeper, why has Alibaba Cloud's profit seen such a significant improvement this quarter?
Perhaps, this is closely related to two factors:
First, profit improvement generated after the increase in MaaS revenue this year.
The massive explosion in the AI Coding market in the first half of this year has significantly increased Token consumption. IDC released its first report on the market share of China's AI programming market in July 2026, showing that the market size of China's AI programming in 2025 was 399 million RMB, and is expected to grow rapidly to 1.173 billion RMB by the end of 2026, with an annual growth rate of nearly 3 times.
Against this backdrop, Alibaba Cloud's MaaS revenue has grown rapidly. Previous earnings reports indicated that Alibaba Cloud's MaaS ARR has exceeded 16 billion RMB, maintaining a state of rapid scaling.
Meanwhile, the next battlefield for AI applications has shifted from coding to work. QwenWork is an enterprise-level AI Agent office platform launched by Alibaba on August 3, 2026. It integrates three products previously launched by Alibaba: QoderWork, MuleRun, and Wukong.
It is reasonable to speculate that by the second half of this year, as MaaS services further scale up, Alibaba Cloud's gross profit may increase further. Currently, NVIDIA's general-purpose GPUs are depreciated over 4 years, while domestically developed cards are depreciated over 3 years. Once the depreciation is complete, it is basically pure profit, requiring only the coverage of electricity and O&M costs.
In other words, as the computing power asset pool gradually enters the "post-depreciation period," the profit release for cloud vendors has just begun. Some institutions have also made relevant forecasts. A Morgan Stanley report shows that with the increasing proportion of MaaS, Alibaba Cloud's profit margin is expected to move from the current 11-12% towards a long-term target of 20%.
Second, full-stack joint optimization further improves efficiency, and this efficiency improvement is ultimately reflected in profits.
It now appears that for the same model, different discounts lead to different profits, and most of the issues stem from the varying degrees of underlying infrastructure optimization. At an investor exchange meeting with Liang Wenfeng a while ago, this issue was also mentioned; there are still differences in infrastructure optimization efficiency between DeepSeek's in-house deployment and cloud vendors' deployment of DeepSeek.
During this earnings call, Wu Yongming also stated that Alibaba has now made considerable collaborative optimizations at the chip, interconnection network, storage, and scheduling layers. For example, the 128-card supernode has increased the intra-rack interconnection bandwidth to the PB/s level, and intelligent scheduling tests can improve model output throughput by 2.5%—5.6% without increasing the amount of hardware.
Third, the accelerated shipment of T-Head's chips can further expand the profit margin.
T-Head's previous-generation chips have cumulatively shipped over 500,000 units. The earnings call also disclosed that as of early August, supernode instances based on the new-generation T-Head chip Zhenwu M890 have been launched for sale on Alibaba Cloud, serving over 650 customers, and shipments will continue to scale up in the second half of the year.
The proportion of T-Head chips in Alibaba Cloud's data centers is also significantly increasing, replacing commercially procured chips. The increase in the proportion of self-developed chips will also substantially boost gross profit margins.
03. Full-Stack AI Layout Brings Composure to Alibaba's AI Commercialization
After reading Alibaba's earnings report, one feeling stands out: compared to single-model companies, Alibaba's full-stack AI approach indeed carries less commercialization anxiety.
Currently, the industry generally divides the AI value chain into four layers: chips, AI cloud, models, and applications.
Over the past two years, market attention has mostly been focused on the model layer. However, as model capabilities steadily improve, a question remains unresolved: exactly how much commercial value can the model itself sustain?
In the long run, when models approach AGI and directly create products and business outcomes in the future, there will certainly be immense room for imagination. But the problem is, before reaching that point, model companies mainly make money by selling Tokens and APIs. Exactly how much revenue and profit this business can support currently remains unanswered.
A few days ago, Bloomberg reported that Anthropic's ARR as of the end of July reached $65 billion, a 7-fold year-over-year increase. Despite a steep growth curve, it still fell short of the institutional expectation of $80 billion. Once the news broke, it directly crashed the AI sector in U.S. stocks that night.
Against this backdrop, at the end of July, another noteworthy signal emerged in the capital market: some funds are repositioning to seek out companies that "can profit from the AI wave regardless of who ultimately wins," causing cloud vendors to surge.
Microsoft is a very typical example. At the end of July, Microsoft released its fourth-quarter earnings report for fiscal year 2026. Azure's single-quarter revenue grew by 43% year-over-year, and its full-year revenue surpassed $100 billion for the first time, a 41% year-over-year increase. More importantly, Microsoft also expects Azure to maintain a 45% growth rate in the next quarter. On the first trading day after the earnings release, Microsoft's stock price rose by 15.5%, marking its largest single-day gain since October 2008, with its market capitalization increasing by approximately $450 billion.
Alibaba is no exception. From early July to now, Alibaba (Hong Kong stock) has risen by over 30%. When the ultimate commercialization outcome for model companies remains unclear, capital votes with its feet.
Alibaba's current full-stack layout, spanning from chips and AI cloud to models and applications, has the advantage that no matter which layer the value of AI flows towards, Alibaba has the opportunity to capture it. In the short term, as AI computing power demand explodes first, Alibaba Cloud reaps the infrastructure dividends first; later on, as the scale of model invocations increases, model revenue can take the baton; further down the line, if AI applications truly achieve scale, the value will continue to migrate towards the application layer.
Moreover, these four layers are not four parallel streams of revenue, but rather an interconnected value chain that drives traffic to one another—applications drive invocations, invocations consume tokens, and tokens consume computing power. The revenue from each layer creates demand for the next.
Therefore, compared to single-model companies, Alibaba, with its full-stack AI layout, may be able to walk the path of AI commercialization with greater composure.