Surge by 889%, exceeding 10 billion yuan in half a year, ByteDance, Alibaba and Baidu are fighting fiercely in the new battlefield.
In 2021, Zhang Xin, founder of CloudCheers, joined ByteDance, which caused quite a stir in the cloud computing circle.
The industry consensus at that time was that ByteDance was preparing to tackle the tough IaaS (Infrastructure as a Service, a cloud computing service model) market, and go head-to-head against Alibaba Cloud, which had occupied the leading position for many years. Zhang Xin's deep experience in container cloud was also regarded by the outside world as a clear signal of ByteDance's determination to build a "complete cloud ecosystem".
But at that time, the IaaS track had seen years of fierce competition, with high moats and narrow breakout windows for latecomers, and Alibaba Cloud, Huawei Cloud, Tencent Cloud, China Telecom Cloud, Baidu Cloud and AWS had almost divided up the entire market.
Over the past five years, the development of the situation has completely exceeded the initial expectations.
On November 12, Zhang Xin, former Vice President of Volcano Engine, left ByteDance. ByteDance's cloud business did not follow the route expected by the outside world to compete in the traditional infrastructure battlefield where Alibaba Cloud had persisted for ten years, but opened up a new growth gap in the MaaS track.
In 2025, the scale of China's AI cloud market reached 56.7 billion yuan, of which the underlying computing power IaaS accounted for 69% of the share, and the model service MaaS accounted for 31%.
The right to speak in the incremental market is being restructured. IDC data shows that in the first half of 2026, the scale of China's MaaS (Model as a Service, the model as a service in the artificial intelligence field) market reached 106.3 billion yuan, a year-on-year surge of 889.4%. Volcano Engine took 45.1% of the revenue share, leading the market by a large margin, nearly 20 percentage points ahead of the second-ranked Alibaba Cloud.
On this new track with explosive growth, Volcano Engine, Alibaba Cloud, Baidu Cloud and Huawei Cloud are brewing a brand new competition order.
01
Entering 2026, MaaS has become a lucrative business
"This year, people around me are switching to MaaS sales, especially in the first half of the year," a cloud salesperson revealed to Tech Planet. In the past, experience in software and hardware sales was enough, but now AI-related background is prioritized, and the core requirement is to be able to deliver performance and bring customer resources.
Many practitioners who have been deeply engaged in the cloud industry for five years have also felt obvious changes: in the first half of the year, major manufacturers concentrated on recruiting MaaS sales staff, and it is not uncommon for traditional cloud sales to transfer to MaaS positions.
Behind the shift in the popularity of sales positions, orders and capital are pouring into the MaaS track at an accelerated pace.
IDC data shows that the scale of China's MaaS market reached 106.3 billion yuan in the first half of 2026, which is the first time that the half-year volume of this track has exceeded 100 billion yuan, with a year-on-year growth rate of 889.4%, making it the fastest growing sub-track in the public cloud sector.
Along with the market explosion is the surge in call volume: the total Token consumption of the whole market in the first half of the year reached 2857.9 trillion, a year-on-year increase of 432.5%.
MaaS has thus stepped out of the developer pilot stage and officially become a standard configuration in enterprise production links.
A year ago, the industry's attitude was completely different. The above-mentioned cloud business salesperson recalled that in 2025, Alibaba Cloud had not yet included MaaS revenue in the formal assessment, "Essentially, at that time, everyone thought it was difficult to make money from MaaS."
In the first half of 2025, the industry was still in the involution period of basic models, with few landing scenarios, and the scale of Token calls failed to increase; superimposed with the white-hot Token price war from the end of 2024 to the middle of 2025, it was difficult to achieve smooth commercialization.
The turning point came in the second half of 2025. Token quotations gradually returned to rationality, large models began to penetrate into all walks of life, and the business model of MaaS gradually became viable.
The focus of major manufacturers has also shifted from simply chasing Token call volume to solid revenue and profits.
Wu Yongming, CEO of Alibaba, has publicly talked about the Bailian MaaS platform for many times. He once said that in the quarter ending in June this year, the ARR (Annual Recurring Revenue) of AI model and application services including Bailian will exceed 10 billion yuan.
According to the disclosure of LatePost, Volcano Engine's MaaS business revenue reached 4.8 billion yuan in the first half of the year, with a gross profit margin of over 55%.
In the eyes of practitioners, the value of MaaS is not limited to its own revenue growth, but it is also becoming an entry point to leverage underlying cloud resources.
"If many customers move to the cloud starting from MaaS services, as model calls and business implementation continue to deepen, they will often purchase basic resources such as computing power and storage synchronously later."
Therefore, what all parties are competing for is not just the Token orders themselves, but the entry point of cloud customers in the AI era. By meeting the enterprise's model call demand, there is an opportunity to bind the entire subsequent cloud resource chain including computing power, storage, deployment and operation and maintenance.
02
An organizational restructuring: MaaS rewrites the combat mode of cloud vendors
In the past, cloud business organizations were built around IaaS and PaaS, and the sales system was divided according to resource types; now it has to make way for MaaS, the organization serves sales, and sales serve AI growth.
Within a week, ByteDance successively announced changes to three technical leaders: Su Jie, head of key technology R&D for Volcano Engine's automotive business, Shi Yunfeng, head of AI chips and DPU, and Zhang Xin, Vice President of Volcano Engine, have successively announced their departures.
In addition to the changes in the management, ByteDance's adjustment intensity is also more thorough among major manufacturers.
In July, ByteDance established the "Creativity Service Platform", merging the Feishu GTM team with the sales, marketing and customer service teams of Volcano Engine, which is under the unified responsibility of Tan Dai, head of Volcano Engine.
Under this new To B sales system, one salesperson can cover the full product lines of MaaS calls, computing power, and Feishu SaaS at the same time, which not only avoids internal customer snatching, but also implements the customer acquisition logic of "taking MaaS as the entry point to drive underlying cloud resources".
At the Feishu Conference on September 15, Liang Rubo, CEO of ByteDance, further clarified to the public that ByteDance will integrate Doubao, Feishu and Volcano Engine into a unified business system, focusing on the integration of Agent, models and enterprise collaboration environment.
At the same time, ByteDance has cut the reporting chain of underlying R&D: the AI chip team is under the management of Yang Zhenyuan, Vice President of ByteDance and head of Volcano Engine business, and Wu Yonghui, head of the Seed large model, reports directly to Liang Rubo.
This arrangement divides short-term commercialization and long-term basic model exploration into two lines: the Creativity Service Platform undertakes the MaaS revenue target and is the current growth engine; the Seed base makes breakthroughs in basic research.
A series of personnel and structural adjustments point to the same choice: ByteDance no longer insists on competing head-on with Alibaba Cloud in the general IaaS battlefield, but transforms Volcano Engine into a delivery base for large model commercialization, and concentrates all To B resources on the incremental track of MaaS and enterprise Agent.
Alibaba Cloud's response logic is "full-stack binding + dual-track sales". On the one hand, it accelerates the iteration of MaaS products and the flexibility of pricing, on the other hand, it deeply integrates AI capabilities with traditional cloud businesses, and participates in competition with full-stack capabilities from self-developed chips, operating systems, databases to large models.
Liu Weiguang, President of Alibaba Cloud Public Cloud, once publicly revealed that now dedicated MaaS sales staff will be assigned to large customers to cooperate with the original IaaS sales staff. He said that even if two or three sales staff overlap for one customer, this cost is acceptable, and the most important thing is not to miss any opportunity.
In March this year, Alibaba established the ATH (Alibaba Token Hub) business group directly led by Wu Yongming, taking Token as the core main line, integrating scattered AI sectors such as Tongyi Laboratory, Bailian MaaS, Qianwen and Wukong, connecting model R&D, service distribution and scenario implementation, and concentrating resources to accelerate MaaS product iteration and commercialization on the premise of retaining the advantages of the original cloud infrastructure.
Baidu Intelligent Cloud has also completed multiple organizational adjustments this year, integrating the MaaS business and Qianfan brand into the Infrastructure Division as a whole, renaming the original Platform Product Division to the Agent Division, and finally forming three major business lines: infrastructure, Agent and intelligent applications.
No matter whether the path is integration or split, the underlying consensus is the same: MaaS is no longer an additional business, but a core variable that forces cloud vendors to reshape their organizations and redefine the way of competing for customers.
03
MaaS reshapes cloud sales: assessment rules have changed, and the price war is still ongoing
All organizational adjustments will eventually fall on the assessment indicators of front-line sales. In the past, the KPI of cloud sales was centered on the revenue of servers, storage and databases, but now this set of rules has gradually become invalid.
Volcano Engine is one of the early vendors that adjusted the assessment orientation. According to the report of LatePost, in 2024, Volcano Engine adjusted the performance assessment indicators of the sales team, putting the sales of MaaS products in the first place. Salespeople can get higher returns from selling MaaS products at the same price than selling traditional cloud services.
Alibaba Cloud's assessment is gradually moving closer to MaaS products. Token call volume and number of paying customers have become important factors, and it no longer takes traditional cloud revenue as the only core.
However, although the assessment standards are iterating rapidly, the real situation of front-line sales staff is not easy.
A salesperson from a leading cloud vendor confessed to Tech Planet that the current MaaS track does not have high thresholds, and the competition eventually evolves into price wars, discount wars and competition for computing power resources. "Mainstream large model customers can basically access multiple platforms, the switching cost is very low, and it is difficult to lock in customers."
Another AI cloud salesperson added that the truly competitive models in the market are concentrated in a few models such as Kimi, DeepSeek and GLM. Some manufacturers will require salespeople to prioritize promoting their own platforms, but whether customers recognize them and are willing to use them is another matter.
Precisely because customer demands are scattered and self-developed models do not necessarily have absolute advantages, in the first half of this year, Tencent Cloud and Alibaba Cloud successively relaxed their policies, allowing salespeople to resell third-party models such as Zhipu, Kimi and DeepSeek.
In contrast, Volcano Engine is in a relatively advantageous position. Seedance has differentiated competitiveness in the video generation track. Some salespeople said that this product can basically be traded at the original price without heavy discounts.
The more realistic pressure comes from the assessment standards. A cloud business salesperson revealed that the assessment indicator on the channel side has shifted from revenue to net profit, which means that the discounts offered by sales at the front line and the price wars they participate in will be directly transmitted to personal performance.
The shift of assessment indicators is essentially the repeated trade-off of vendors between high-speed growth and profit bottom line. The old revenue-oriented rule encourages the expansion of revenue scale, while the new net profit assessment requires salespeople to learn to calculate costs and profits.
In the stage where MaaS has not yet formed a stable profit model, front-line sales staff are the most direct bearers of pressure in this trade-off.
This article is from the WeChat official account "Tech Planet" (ID: tech618), written by Ren Xueyun, and authorized for release by 36Kr.