Alibaba cannot afford to lose the battle in AI.

This is not the last war between Alibaba and ByteDance, but it is likely to be the most crucial one.

In 2016, Jack Ma put forward a judgment within Alibaba: "AI is not just a technology, but a form of production relations." In his view, the value of technology lies in integrating into the business ecosystem and changing transaction efficiency. In the same year, Zhang Yiming repeatedly emphasized another point within ByteDance: "Algorithms are not an appendage of products, but the core of products." He believes that technology itself can be a product and become a new underlying operating system.

In the following decade, the two corporate cultures collided repeatedly. In the e - commerce battlefield, Alibaba first built an online mall, established payment and logistics systems, creating an ecosystem before attracting traffic. ByteDance developed Douyin, monetizing the attention of a large number of users. In terms of their goals, Alibaba aims for GMV, while ByteDance focuses on user time, and the results of the competition vary.

Now, the war has spread to the AI cloud market. The two companies are vying for the top position in the segmented fields of the AI cloud. Alibaba no longer has the ease of having parallel battle - fronts, and ByteDance is no longer just an intruder. This is not a simple story of a newcomer challenging an established giant.

The global narrative of Google Cloud's counter - attack against AWS with AI provides an important reference for this war. The competition between ByteDance and Alibaba in the AI cloud market will determine the power structure of the Chinese Internet in the next decade.

01

One War, Two Rankings

If there is anything that has most troubled the management of Alibaba Cloud in the first half of 2026 in the Chinese AI cloud market, it is that Alibaba Cloud, the leader in the AI cloud business, was outperformed by Volcengine by 20 percentage points in terms of token call volume, the most telling indicator of the market's "voting with feet".

"Being sniped by Volcengine in the MaaS market revenue is the most painful thing for Alibaba this year," an insider close to Alibaba told us. Although Alibaba Cloud announced at the beginning of the year that its goal was to capture 80% of the incremental market in the Chinese AI cloud market in 2026, in fact, in the AI cloud market driven by strong model capabilities, Alibaba Cloud was outperformed by Volcengine of ByteDance.

Alibaba Cloud has sold the most AI infrastructure, but many enterprise users are betting their future application ecosystems on Volcengine.

It is not difficult to deconstruct the current Chinese AI cloud market. In the AI cloud market with a total scale of 56.7 billion yuan in 2025, IaaS, which sells computing power, accounts for about 69%, and MaaS, which sells intelligence, accounts for about 31%. In terms of revenue, Alibaba Cloud is undoubtedly the leader in cloud computing, with a scale larger than the sum of the next three competitors.

This year, Alibaba Cloud and Volcengine each have a report from a third - party authoritative institution. According to the Omdia report, in terms of the overall AI cloud revenue covering the entire IaaS - PaaS - MaaS chain, Alibaba Cloud ranks first with a 38.1% share, followed closely by Volcengine with 20.4%.

However, the IDC report shows that in the Chinese enterprise - level MaaS market in 2025, Volcengine ranked first with a 49.5% token call volume, and Alibaba Cloud ranked second with 28%. It should be noted that in 2025, the call volume of large models in China's public cloud soared to 194.4 trillion Tokens, a 16 - fold increase year - on - year.

In this rapidly expanding incremental market, Volcengine maintained its 49.2% share in the first half of the year without dilution and firmly held the top position in token calls.

This was a successful surprise attack. ByteDance bypassed Alibaba's traditional IaaS market and penetrated with lightweight API services. By the time the opponent reacted, it had already secured a position in the incremental market.

(In the traditional cloud - computing era, cloud services include IaaS (Infrastructure as a Service), which provides servers, storage, and networks; PaaS (Platform as a Service), which provides development frameworks and middleware; and SaaS (Software as a Service). In the AI era, a fourth layer, MaaS (Model as a Service), has emerged on top of these three layers. It packages pre - trained large models into standardized API interfaces. Users can obtain AI capabilities by calling these interfaces without training models themselves and are billed according to token consumption.)

This is not the first time Alibaba has faced such a situation.

Ten years ago, Taobao defined e - commerce with product categories and logistics, but ByteDance took away users' time with Douyin, and brand merchants had to open stores on Douyin. Ten years later, Alibaba Cloud defined the cloud - computing market with full - stack capabilities and government - enterprise relationships, but Volcengine captured customers' minds with Doubao and models.

The logic is exactly the same. In the current era where "attention is all you need", ByteDance is still using new - paradigm traffic entrances to capture transaction shares from the old paradigm. This logic was verified in e - commerce and is now being reused in the AI cloud market.

Interestingly, Alibaba has not lost in every battle but has been penetrated by ByteDance with a different business logic. Now, it's the turn of the AI cloud.

It is in this dimension that Volcengine has taken the lead.

"This year, cloud providers have all ridden on the wave of the lobster token craze, but Volcengine has advanced faster than everyone else. Model capabilities are the key behind this," Cheng Ren, a post - training technician from a leading large - model manufacturer, told us. If Alibaba's models are strong in pre - training and well - known in the open - source ecosystem, ByteDance's models are more like all - around players.

From the Seedance 2.0 that went viral during the Spring Festival, the Seedream 5.0 focused on image generation, to the voice - generation model that was called the world's best by Google DeepMind AI scientist Yao Shunyu, ByteDance has comprehensively expanded its multimodal product line. For AI application entrepreneurs, the most fundamental survival rule is to use the latest and best models to retain users, and everyone is racing against time.

The usability of the model directly determines how fast and far the product can go.

In the eyes of industry insiders, the turning point for Alibaba's large models began with an important personnel change in the summer of 2024. At that time, Zhou Chang, the technical leader of Tongyi Qianwen, left Alibaba and joined ByteDance's Seed team, along with more than a dozen key team members under him. Zhou Chang then led ByteDance's multimodal models to surpass others, and Seedance 2.0 was developed by his team.

At Alibaba, Zhou Chang was always a key figure in model iteration. He not only led the research and development of the multi - trillion - parameter multimodal pre - trained model M6 series (the predecessor of the Qwen large model). In early 2024, the Qwen 2.5 series gained the reputation of "China's strongest open - source model" internationally.

At that time, the model was entering a critical period that required core figures to make technical judgments on the direction. Zhou Chang's departure was considered to have affected the technical continuity of Alibaba's model iteration. After Zhou Chang left, Lin Junyang took over the Tongyi Laboratory, but he was not the actual technical leader of the Tongyi Qianwen large model. Two years later, Lin Junyang, the post - training leader Yu Bowen, and core member Li Kaixin left one after another. The core technical layer of Tongyi Qianwen experienced two rounds of key personnel departures in less than two years.

In April, Alibaba announced the upgrade of the Tongyi Laboratory to the Tongyi Large - Model Division, which is now led by Zhou Jingren, the CTO of Alibaba Cloud. At the group level, a technology committee was established, with CEO Wu Yongming personally serving as the leader, and the members covering the entire chain of models, computing power, and engineering. However, unfortunately, the Qianwen team, which should have been in a coordinating position, was not responsible for the technical exploration and development of multimodality. The Wanxiang Laboratory responsible for multimodality is under the management of Alimama under Taotian Group.

The spill - over effect of Alibaba's talent is also evident in some specific scenarios.

Since December 2025, Alibaba has been going through a trough in the Vibe Coding, the most active developer scenario in the AI era. When the Vibe Coding developer community refers to GLM - 4.7 as a substitute for Claude, the voice of the Qianwen community has weakened due to the departure of core members.

"Although Alibaba's Qwen Code has been upgraded from a tool to a full - stack platform with the release of v0.5.0, it still seems half a step behind. In the current situation where the iteration cycle of large models is calculated in months, the window period for Alibaba's models to make up for the gap will not be long," Cheng Ren added.

02

Why the AI Cloud Will Be the Final Battle

When Tan Dai was interviewed by LatePost, he gave Volcengine a "M +" rating, which was partially beyond expectations. However, what he really wants is an "O" (excellent level), meaning "everyone will choose Volcengine when it comes to AI". Even when Volcengine already occupies nearly half of the token call volume, his judgment is still that "the marathon has only run 500 meters".

This statement is very much in line with Zhang Yiming's style, always feeling that one is not fast enough and never being satisfied. Tan Dai's expectation that "everyone will choose Volcengine when it comes to AI" is exactly the source of Alibaba's anxiety.

In previous battles, Alibaba had a fallback. If it lost some shares in e - commerce, it still had the cloud business. If it burned some money in local life services, it still had Ant Group. But there is no fallback in the AI cloud market. The cloud is the infrastructure entrance in the AI era, the fastest - growing business segment of Alibaba, and the core anchor for the capital market to re - evaluate Alibaba.

In the e - commerce battlefield, merchants can open stores on both Taobao and Douyin, selling the same inventory on multiple platforms. Switching platforms is more like adjusting the focus of investment, with almost zero cost. However, in the cloud - service market, migration is much more painful than expected.

Once a company builds all its core business data, models, deployment tools, and monitoring systems on a particular cloud, with databases, middleware, monitoring systems, and operation and maintenance scripts all built based on the proprietary APIs and toolchains of that cloud. Running this system on another cloud means rewriting it from scratch. This means that once a customer chooses a particular cloud, its business may be locked in for a long time.

Now, the AI story of Alibaba Cloud is becoming the core narrative to support the market value of the entire Alibaba Group, and the MaaS call volume has become a glaring footnote.

According to Alibaba's just - released Q4 financial report for fiscal year 2026, the external commercial revenue of Alibaba Cloud accelerated to 40%. The quarterly revenue of AI - related products reached 8.971 billion yuan, achieving three - digit growth for 11 consecutive quarters, and the annual recurring revenue exceeded 35.8 billion yuan. This is the first time Alibaba has clearly disclosed the proportion of AI revenue in its financial report and the first time it has disclosed the annual recurring revenue (ARR) of AI - related products.

Wu Yongming clearly predicted in a conference call that the proportion of AI - related product revenue will exceed 50% in the next year, and the ARR of AI models and application services, including the Bailian MaaS platform, will exceed 30 billion yuan by the end of the year.

The more acute problems are hidden in the details.

IDC data shows that in terms of the token call volume of external customers, Volcengine's share in the whole year of 2025 was 49.5%, up from 49.2% in the first half of the year. Alibaba Cloud's share was about 27% in the first half of the year and about 28% for the whole year.

The explosive growth of annual tokens has masked the fact that Alibaba's expansion speed in the incremental market is not enough to shake Volcengine's leading position. The opponent's share is not shrinking but strengthening.

Of course, there is another way to calculate. In the Omdia's full - chain AI cloud revenue ranking, Alibaba Cloud ranked first with a 38.1% share, proving that its dominance in the IaaS layer remains solid. The growth of Alibaba Cloud's AI product revenue mainly comes from embedding AI capabilities into existing large customers, such as helping financial institutions build risk - control models, helping car manufacturers with autonomous - driving simulations, and helping the government with data governance.

These high - value workloads often involve private deployment and are not reflected in token call volume. However, in the Omdia report, in the Chinese AI cloud market in H1 2026, Baidu jumped to the second place with a 26.9% share, while Alibaba Cloud's share dropped from 38.1% to 35.8%. Although Alibaba still ranks first in the Omdia overall ranking, its share has decreased in the context of the expanding market.

The problem is that in the Chinese AI cloud market with a total scale of 56.7 billion yuan in 2025, IaaS accounts for about 69%, and MaaS accounts for about 31%. The token call volume ranking measures the 31% that has a faster growth rate and better represents the future.

To some extent, the global battlefield has provided a reference.

In Q1 2026, Google Cloud's revenue reached $20.03 billion, a 63% year - on - year increase, far exceeding AWS (Amazon Web Services)' 28% and Azure's 28%. Amazingly, Google Cloud's backlog of orders soared from $240 billion in the previous quarter to over $460 billion, almost doubling. Although AWS still ranks first with a 28% global share, it has dropped one percentage point from 29% in Q1 2025.

Many analysts pointed out that Google Cloud's growth is almost entirely driven by AI workloads. Its strategy focuses on AI - native projects, while AWS and Azure's attention is still occupied by a large amount of traditional cloud business (legacy footprints), which is why the former has a faster growth rate.

Applying this to the domestic market, Alibaba Cloud is like AWS, with the largest existing volume and the thickest foundation. However, its growth depends on adding AI capabilities to traditional customers rather than establishing dominance in purely new AI scenarios. As a reference, Volcengine is more like Google Cloud, entering from the AI - native MaaS entrance, bypassing the existing barriers of traditional IaaS, and rewriting the pattern with the incremental market.

The rise of Google Cloud proves that in the AI - driven cloud era, technical capabilities are more important than market - share inertia. If Google can increase its backlog of orders from $157.7 billion to $240 billion in one quarter, then Volcengine's goal of achieving tens of billions in MaaS revenue is just a starting point.

This potential growth curve is the real source of Alibaba's anxiety.

In China, Volcengine plays the role of a challenger with AI - native genes. Omdia predicts that the scale of the Chinese AI cloud market will reach 193 billion yuan by 2030. This is a rapidly expanding new market, large enough for any latecomer to overtake others.

For Tan Dai and Volcengine, they locked in the incremental space of this war early with a product - oriented mindset. Volcengine let customers experience the effects of the product (Doubao) first and then sold the same capabilities in bulk. This is a classic strategy that ByteDance verified in the e - commerce era, using a large number of users to spread the cost and then using the cost - advantage to crush the competition.

A cloud - business practitioner revealed that since 2024, Volcengine has adjusted the assessment indicators for its sales team, with MaaS products ranked first. The rewards for salespeople for selling M