After years of heavy investment, returns are beginning to show up for Alibaba.
The Chinese internet giant’s revenue for the fourth quarter of fiscal 2026, which corresponds to the first calendar quarter, reached RMB 243.38 billion (USD 35.7 billion), up 3% year-on-year (YoY). Excluding revenue from divested businesses such as Sun Art Retail and Intime, comparable revenue grew 11% YoY.
The double-digit comparable revenue growth was driven mainly by Alibaba Cloud. The cloud business continued to grow quickly in the first quarter, with revenue up 38% YoY, broadly in line with market expectations.
Revenue from external customers grew by more than 40% YoY, accelerating from 35% in the previous quarter. AI-related product revenue recorded triple-digit growth for the 11th consecutive quarter, while continued growth in public cloud also contributed to the segment’s performance. E-commerce revenue was comparatively stable, with domestic and international e-commerce each growing 6% YoY.
Alibaba’s profitability weakened amid its current transition phase. The company recorded an operating loss, while adjusted EBITA fell 84% YoY to RMB 5.2 billion (USD 763.4 million) from RMB 32.6 billion (USD 4.8 billion) in the same period last year. Non-GAAP net income fell to RMB 86 million (USD 12.6 million) from RMB 29.85 billion (USD 4.4 billion) a year earlier.
However, gains from AI-related investments, including MiniMax and Z.ai, lifted Alibaba’s investment returns. Interest and investment income, net, reached RMB 87.5 billion (USD 12.8 billion) in fiscal 2026, up 322% YoY and its highest level since 2022. The gain reversed a period in which weaker markets had weighed on many of Alibaba’s investments.
Since the start of this year, the arrival of agentic artificial intelligence, meaning AI systems that can plan and execute tasks with limited human input, and surging demand for computing power have triggered a global rally in AI-related stocks. Sentiment has also been boosted by US President Donald Trump’s visit to China with executives from several major US companies. Yet Alibaba and Tencent, both of which have invested heavily in AI, have delivered weaker share price performances.
To stabilize market sentiment, Alibaba CEO Eddie Wu offered a clearer answer after previously setting a target for AI and cloud revenue to reach USD 100 billion annually within five years. On the earnings call, he said that, based on surging global demand for model-as-a-service (MaaS), the ARR (annual recurring revenue) of Alibaba’s AI cloud business, namely Model Studio’s annualized revenue from AI model and application services, will exceed RMB 10 billion (USD 1.5 billion) by June and surpass RMB 30 billion (USD 4.4 billion) by year’s end.
Wu also expects AI-related revenue to account for more than 50% of Alibaba Cloud’s revenue over the next year, making it the cloud business’s main growth engine. Alibaba Cloud already generates more than RMB 100 billion (USD 14.7 billion) in annual revenue, so a 50% share would imply tens of billions of RMB in incremental revenue.
Alibaba turns toward AI revenue
As one of the first major Chinese tech companies to declare an all-in AI strategy, Alibaba has begun to talk more about commercialization over the past two quarters, rather than simply emphasizing spending. At the start of the earnings call, Wu disclosed specific AI revenue figures for the first time: Alibaba Cloud’s AI-related product revenue for the quarter was RMB 8.971 billion (USD 1.3 billion), with annualized revenue exceeding RMB 35.8 billion (USD 5.3 billion) and accounting for 30% of Alibaba Cloud’s revenue. Three quarters earlier, the figure was 20%.
While maintaining a three-year spending plan of RMB 380 billion (USD 55.8 billion), Alibaba’s current capital expenditure level is more stable than that of Tencent and ByteDance. Its first-quarter capital expenditure was RMB 26.8 billion (USD 3.9 billion), broadly flat with RMB 24.8 billion (USD 3.6 billion) in the same period last year. Tencent said it aims to double capital expenditure this year, and its first-quarter capital expenditure reached RMB 31.9 billion (USD 4.7 billion), significantly higher than RMB 19.6 billion (USD 2.9 billion) a year earlier. ByteDance has also recently raised its capital expenditure again to RMB 200 billion (USD 29.4 billion), more than double last year’s total.
The difference mainly reflects the companies’ different stages and development paths. Like Google, Alibaba moved early to build a full-stack model spanning investment, infrastructure, models, and deployment. Tencent’s significant increase in AI spending seemed more like a catch-up effort. At Tencent’s shareholder meeting on the day it released first-quarter results, CEO Pony Ma acknowledged that the company’s early AI foundations were not especially strong. He said Tencent once thought it had secured a place in the AI race, only to realize later that its footing was still unstable.
Tencent is now increasing investment in large models and WeChat agents. Its approach is to embed AI functions into the existing WeChat ecosystem, rather than build a fully integrated platform from scratch, as Alibaba has done.
ByteDance’s capital expenditure is markedly higher than that of other vendors. This is not only because ByteDance is also moving toward a full-stack layout, but also because the future high-speed growth of Volcano Engine will depend heavily on ByteDance’s investment in video models. These require far more computing power than large language models, naturally driving costs higher.
The giants’ capital expenditure, often in the hundreds of billions of RMB, must eventually translate into commercial returns. Compared with US peers, however, China’s overall AI sector still lacks the monetization efficiency needed to support long-term investment at the RMB 100 billion level. Although the gross margin of AI cloud, at 60–80%, is much higher than that of traditional cloud, at 15–20%, China’s AI cloud business faces a price war that is unlikely to end soon.
Meanwhile, DeepSeek’s V4 model directly offered a limited-time 75% discount, forcing vendors to confront the reality that net profit margins are unlikely to improve significantly for now. Accordingly, Wu’s previous target for Alibaba Cloud’s net profit margin to reach 20%, roughly doubling from the current level, will still take time.
Wu also said directly on the earnings call that Alibaba’s AI business is entering a cycle of commercial returns. Increasing AI revenue is Alibaba’s top priority at present. Given current industry conditions, profit margin can only come second. Alibaba Cloud’s AI revenue mainly comes from proprietary models, much of it from API services based on Qwen models and MaaS. A smaller portion comes from subscriptions to AI-native software such as DingTalk.
If traditional cloud computing is Alibaba Cloud’s base business, AI cloud is the next growth area it needs to capture. It could shape Alibaba Cloud’s long-term growth potential and determine whether Wu’s target of USD 100 billion in annual AI and cloud revenue, spanning AI cloud and traditional cloud services, is achievable.
Citi forecasts Alibaba’s AI and cloud revenue will rise sharply over the next five years, with a compound annual growth rate of 90%, while MaaS revenue growth could exceed 200%. The forecast is built around a new commercial flywheel: merchants use Wukong to improve operational efficiency, and Taotian Group merchants could become Wukong users, creating new revenue opportunities.
Wu said Alibaba Cloud is shifting from traditional cloud computing to model computing and agent services. Since the AI boom began, the traditional model of selling storage, servers, and bandwidth has increasingly given way to selling model calls, computing power, and intelligent services. This means Alibaba Cloud must break from some of its past operating habits. Alibaba Cloud has dominated China’s traditional cloud market for many years and once competed head-on with powerful rivals such as Huawei Cloud. But it still faces significant pressure from ByteDance, an aggressive rival in AI cloud.
With the goal of selling as many tokens as possible, Alibaba is internally positioning Model Studio as the main vehicle for its MaaS business.
36Kr has learned that Model Studio’s revenue last year was still limited, which would make Wu’s RMB 30 billion forecast a major leap. Model Studio currently integrates more than 200 mainstream third-party large models, including Qwen, DeepSeek, Kimi, and MiniMax. More than one million companies and individuals have reportedly used it, and more than 800,000 AI agents have been created. In the first quarter, Model Studio’s customer count grew eightfold.
ByteDance is likewise expanding its AI cloud business. It is helping Volcano Engine generate revenue by selling video models such as Seedance and has set more aggressive targets. 36Kr has learned that Volcano Engine raised its MaaS revenue target for this year to RMB 15 billion (USD 2.2 billion), up from RMB 10 billion at the start of the year. The higher target suggests ByteDance is also finding additional commercialization paths for its AI business.
To compete for a larger market share, Alibaba needs to keep developing different types of models for different fields, including coding, multimodal video generation, and future-oriented world models, which are AI systems designed to simulate environments and predict outcomes. This still requires continued investment in talent and technology. One person close to Alibaba told 36Kr:
“Model iteration is too fast. A top model may change every six months. Only by constantly developing new models is it possible to catch new trends and hot spots.”
36Kr previously learned that, to sell more tokens, multimodality will also be one of Alibaba’s key model-level development directions this year.
Recently, Alibaba also spent money to acquire a research institute specializing in multimodal technology. HappyHorse, launched recently, is the first video generation app released by the institute after it was integrated into the Alibaba Token Hub business group. It is currently trying to win users from Seedance and Kuaishou’s Kling AI by relying on advantages such as cost-effectiveness, generating videos of the same length while consuming fewer tokens, and more efficient video generation. Its commercialization is later than that of peers, with ordinary users being prioritized over video companies.
Alibaba is not only trying to generate revenue, but also improve the cost structure of its AI business. Tight computing supply is a shared challenge for major technology companies. By extension, producing in-house chips at scale could be an important way for Alibaba to control costs.
For reference, the strategy of using in-house development to reduce costs has already played out successfully in the US. Amazon’s in-house AI chip business, Trainium and Inferentia, has surpassed USD 20 billion in annualized revenue and is growing at a triple-digit pace. This shows that Amazon is trying to reduce its dependence on Nvidia. Alibaba’s chip subsidiary, T-Head Semiconductor, has also achieved scaled mass production and broad industrial application of in-house GPU chips. More than 60% of its computing capacity serves external commercial customers, covering use cases across the internet, finance, and autonomous driving sectors.
According to information 36Kr obtained from institutional sources, Alibaba’s capital expenditure in 2026 will rise to RMB 150–170 billion (USD 22.0 billion–25.0 billion). This will depend largely on how many H200 chips it can obtain, which in turn will depend on geopolitical conditions. Roughly half of the spending is reserved for chip procurement, with around RMB 30 billion allocated to domestic chips. Most of that budget will be converted into T-Head’s Zhenwu 810E.
In addition, 36Kr has learned that the widely rumored financing talks between DeepSeek and Alibaba actually took place last year. DeepSeek initiated the talks, saying it wanted access to scenarios such as Alipay, while Alibaba also needed DeepSeek’s infrastructure capabilities. However, the talks did not continue for various reasons. The trigger for renewed cooperation this year was discussions between T-Head and DeepSeek over chip procurement.
Wu said on the earnings call that almost all AI compute chips in Alibaba’s servers are being used efficiently, with no resource waste. The comment rebutted the argument that there is a compute oversupply and emphasized that Alibaba’s AI business has entered a stage of scaled returns.
Qwen and Taobao test an ecosystem play
After a year of intense competition, Taobao Shangou (also known as Taobao Instant Commerce) has a new mission and target. 36Kr previously learned that Shangou’s target for the new fiscal year is to achieve positive unit economics while maintaining market share. Executives made a corresponding commitment on the earnings call, meaning the phase of expanding quick commerce through subsidies has ended.
Given that heavy spending on two fronts has damaged profits, Alibaba’s e-commerce business needs to spend more cautiously. In the first quarter, adjusted EBITA for Alibaba’s China e-commerce business fell 40% YoY, with losses of more than RMB 20 billion (USD 2.9 billion). Excluding losses from the quick commerce business, EBITA was stable YoY. In addition, first-quarter cash flow turned negative again, with heavy investment causing a net outflow of RMB 17.3 billion (USD 2.5 billion).
Taotian Group’s customer management revenue (CMR) has returned to normal growth. Under the new reporting basis, the e-commerce business launched a plan to return part of marketing expenses to merchants if they meet gross merchandise value (GMV) targets. Previously, merchants’ advertising fees and commissions were included in CMR, while marketing expenses were included in market spending. In the first quarter, China e-commerce CMR grew 1%, but on a comparable basis, actual YoY growth was 8%, exceeding market expectations. This was mainly due to delayed Lunar New Year stockpiling and continued revenue growth from Taobao Shangou.
Taobao Shangou’s path to positive unit economics depends on continually reducing costs and expanding merchant revenue.
At present, Taobao Shangou has made reasonable adjustments to both user and rider subsidies. After a year of user accumulation, Alibaba’s subsidy focus has shifted toward users with higher average order values. Rider subsidies have also been adjusted based on order volume and business layout. Compared with last year, more Shangou rider capacity will support retail businesses such as four-hour and half-day delivery.
Some leading merchants now have a strong willingness to pay because early adopters have already seen results. Take Decathlon as an example. Since launching a direct sales plus in-store pickup model on Taobao Shangou, its average daily order volume has reportedly exceeded 5,000, and peak order volume during major promotions can reach 20,000. Its single-store fulfillment radius has expanded from three kilometers to ten kilometers, and sales have already surpassed RMB 100 million (USD 14.7 million).
This is closely tied to Taobao Shangou’s push into retail. Since the end of last year, Shangou has expanded its retail footprint, increasing investment in formats such as warehouses built for rapid local fulfillment, Tmall Supermarket, and Hema front warehouses. A person close to Shangou told 36Kr that fresh groceries still account for 70–80% of its retail business, but its target this year is to increase the combined share of daily necessities, beauty products, alcoholic beverages, and healthcare products to about 50%. The SKU count for each of these categories will increase by more than 1,000.
The next step is to make those retail categories easier to discover and convert. That is where Qwen enters the e-commerce story. In the same week Alibaba released its earnings, it announced that Qwen had been integrated into the Taobao app across all e-commerce categories. Taobao also launched an AI shopping assistant powered by Qwen, which will provide services such as outfit styling and price comparison. The integration came only one quarter after Qwen was connected to Taobao Shangou, making it a relatively fast rollout compared with other AI e-commerce efforts globally.
36Kr has learned that Alibaba’s approach to consumer-facing AI differs from the more sales-driven orientation of AI in the business-facing market. For B2C use cases, Alibaba will evaluate token consumption, user volume, and ecosystem synergy together. Rather than focusing only on consumer user numbers, it is also examining how products such as Qwen can coordinate with other business segments inside Alibaba’s ecosystem.
The main external question around the integration of e-commerce and Qwen is whether Alibaba can succeed where major US technology companies have struggled.
In AI-driven e-commerce, Alibaba’s advantage is that it has both a leading open-source large model family, Qwen, and a sizable e-commerce platform, Taobao. It can pursue deeper integration without cross-company coordination. This is also why, after Alibaba’s market value was overtaken by Pinduoduo in 2023, Jack Ma could write confidently on Alibaba’s internal forum:
“The era of AI e-commerce has just begun. It is an opportunity and also a challenge for everyone.”
Major US tech companies have faced a structural constraint: few possess both advanced large model capabilities and a large e-commerce ecosystem. OpenAI and Google do not have e-commerce ecosystems, while Amazon has lagged behind leading AI model developers. To explore new opportunities and expand its business-facing services, OpenAI turned toward Amazon and loosened its tie with Microsoft, shifting to a nonexclusive partnership.
But the window may not stay open for long. ByteDance is putting pressure on Alibaba from the other direction, through Doubao and Douyin E-commerce, whose integration is expected to deepen. For Alibaba, the challenge is not just improving Taobao with Qwen, but doing so before ByteDance narrows the gap.
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Peng Qian for 36Kr.
Note: RMB figures are converted to USD at rates of RMB 6.81 = USD 1 based on estimates as of May 21, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.