Header image generated by AI

Text | Entropy Field

Editor | Shen Xiao

In July this year, Wang Teng, former general manager of Xiaomi China's marketing department and general manager of the REDMI brand, posted photos on Weibo of airport billboards from two cloud vendors.

He wrote: "In the AI era, airport billboards are already plastered with cloud vendor ads. One says its share exceeds the combined total of the industry's 2nd to 4th place, and another says its market share is 49%. Based on my industry experience, neither is probably lying—it's just that the data scopes are different."

These two billboards came from Alibaba Cloud and Volcano Engine. By the end of October, both companies had produced new results.

In the China full-stack AI cloud market as measured by Frost & Sullivan, Alibaba Cloud ranked first in the industry with a 42% market share, exceeding the combined total of the second to fourth places.

Several hours later, in the "2026H1 China AI Software Market Tracker Report" released by IDC, under the MaaS statistical scope, Volcano Engine took the top spot in domestic general-purpose model service revenue with a 45.1% share.

MaaS mainly covers large language models, image models, and video models, corresponding to the three content modalities of text, images, and video.

Under this broad scope, Volcano Engine held the lead through its video model business; Alibaba Cloud, with a 46.6% market share, took first place in the large language model sub-market.

Snatching the large language model crown from Volcano Engine, which ranks first in overall MaaS revenue, was a case of Alibaba Cloud wresting food from the tiger's mouth. Some industry insiders believe that Alibaba Cloud's ability to take first place in large language models is related to Alibaba's past investments in native large model startups such as Kimi.

Volcano Engine's growth, by contrast, comes more from ByteDance's productization capabilities in the Seedance series of models. Video generation models have entered content production scenarios such as short dramas, also bringing Volcano Engine faster demand growth.

The seemingly intensifying battle for first place is in fact undercurrents surging before the large model competition has reached its endgame.

After all, in the AI era where models determine everything—whether selling computing power, selling chips, or selling cloud services—a usable model is the biggest variable for explosive market demand.

Where the model is, that is where the market's number one will be.

The ranking hierarchy in the cloud market was once strictly stratified.

Rewind to 2023, the first year of domestic generative AI commercialization. Enterprises were still mainly buying traditional cloud services. In this foundational business, market position depended mainly on revenue from basic cloud resources such as computing and storage, and Alibaba's lead was relatively clear.

According to Canalys, in the fourth quarter of 2023, in the domestic cloud infrastructure services market, Alibaba accounted for 39%, Huawei 19%, and Tencent 16%, with the three together taking about three-quarters of the market.

As for the AI cloud market, the industry did not yet have a refined statistical system at the time. The so-called "AI public cloud services" mainly counted existing AI businesses before the rise of large models, such as computer vision, speech, natural language processing, and machine learning.

In March of that year, Baidu's ERNIE Bot was the first to begin internal testing, and third-party statistical agencies began to measure market position by combining comprehensive large model capabilities and test invocation volume.

According to IDC statistics, the domestic AI public cloud market size in 2023 was 12.61 billion yuan. In August 2024, Baidu cited this report and announced that it had ranked first for five consecutive years with a 26.4% share.

In November of the same year, a share chart reposted by Shanxi Securities International showed that Alibaba's share was also 26.4%, catching up with Baidu. Tencent and Huawei stood at 20.7% and 17.8%, respectively.

So at that time, in selling basic cloud services such as computing power and storage, Alibaba led; in selling AI services such as speech recognition and image recognition, although Baidu held the "first for consecutive years" banner, Alibaba's share had already caught up. Tencent and Huawei were also at the top in both types of markets.

Volcano Engine was still chasing from behind.

It only officially released its full range of cloud products in December 2021, launching 78 services across five major categories at once. At the time, Volcano Engine general manager Tan Dai told the media that the goal was to become an important participant in the entire cloud market. He emphasized that enterprises usually use multiple clouds, which leaves opportunities for latecomers.

Large models provided that opportunity.

In 2024, domestic AI application startups ushered in an explosive period, with more than one million AI mini-programs and application tools launched throughout the year, and market-based invocation volume of general-purpose models rising 470% year on year.

Cloud vendors such as ByteDance and Alibaba, which themselves have giant internal AI businesses, rose rapidly. Internal computing power consumption grew larger and larger, even catching up with external market-based traffic. The track saw a split in statistical scope, and third-party agencies added a new core distinguishing criterion: whether to count enterprises' internal self-use computing power traffic.

On May 15, the Doubao large model was officially sold to external enterprises through Volcano Engine. The input price of the Doubao general model Pro-32K was set at 0.8 yuan per million tokens.

Six days later, Alibaba announced that the input price of Qwen-Long would be reduced to 0.5 yuan per million tokens; Baidu announced on the same day that its ERNIE Speed and ERNIE Lite models would be free. On May 22, Tencent also adjusted the pricing of its Hunyuan models, with Hunyuan Lite becoming free.

The new model business brought new rankings.

An IDC report released in April 2025 showed that in the token invocation volume (MaaS) of China's large model public cloud services in 2024, Volcano Engine accounted for 46.4%, while Alibaba and Baidu each accounted for 19.3%. For the first time, Volcano produced a national share sufficient to surpass the old vendors.

By 2025, multiple institutional reports from Frost & Sullivan, IDC, Omdia, CCID, and others were released, and leading vendors each obtained first place under different yardsticks.

In Frost & Sullivan's full-stack AI cloud overall market, Alibaba Cloud ranked first with a 40.1% share, while Baidu and Volcano followed with 16.6% and 13.4%, respectively.

IDC's statistics on token invocation volume of public cloud large models showed Volcano Engine leading with 49.5%, Alibaba at 28%, and Baidu at 10%. A year earlier they were in the same position; a year later, Alibaba was already nearly three times Baidu.

Switching to AI application public cloud service revenue, Baidu ranked first with 30.7%. Further segmented, in the 238 million yuan online media processing service and the 155 million yuan audio-video AI real-time interaction market, Tencent took both first places.

Huawei, meanwhile, continued to emphasize its achievements in industry cloud. Citing an IDC report, Huawei announced that Huawei Cloud Stack ranked first in China's financial private cloud infrastructure market in the first half of 2025, with a share higher than the combined total of the second to fourth places.

Since then, the AI cloud market has bid farewell to the era of a "unified champion." A single report can mass-produce multiple number ones, and ranking no longer represents comprehensive strength but rather adaptability under a single set of rules.

The fight for first place is a ticket to revenue and profit in the AI era. The giants all want to seize the window period and achieve big results.

In the second quarter of this year, Alibaba Cloud's external commercialization revenue grew 45% year on year, and AI-related product revenue achieved triple-digit growth for the twelfth consecutive quarter; adjusted EBITA for AI cloud and computing power services was 5.628 billion yuan, up 133% year on year.

In the same quarter, Alibaba's China e-commerce business revenue was 110.9 billion yuan, down 8% year on year; customer management revenue fell 7% year on year, and grew 1% on a comparable basis after excluding the accounting impact of new marketing plans. The cloud business is accelerating, while the e-commerce business that once provided traffic and profit is under pressure.

Baidu faces more urgent changes.

In the second quarter of 2026, traditional business revenue was 10.4 billion yuan, down 23% year on year; core new AI business revenue was 12.5 billion yuan, up 25% year on year, accounting for half of general business revenue. Among this, AI cloud infrastructure revenue was 7.3 billion yuan, up 50% year on year, and GPU cloud revenue grew 283%. But the incremental new business is not yet enough to offset the contraction of the old business.

ByteDance's situation is different. Advertising, e-commerce, and local life are still providing growth. As a latecomer to the cloud market, Volcano Engine, relying on its massive C-end business, aims to export its self-developed models and computing power capabilities to win more enterprise customers.

Model service is a more mature business model than selling tokens, with a lower threshold for enterprise access, and invocation volume rankings happen to provide visibility for this path as well.

Volcano Engine president Tan Dai said in an interview in June this year that before the launch of Seedance 2.0, video generation invocation volume was usually higher on weekends; after the launch, weekday invocation volume clearly exceeded weekends. As of then, the number of customers with annual cumulative token usage exceeding one trillion had surpassed 200, double the number at the end of 2025.

Tencent has a different set of books. In the second quarter of 2026, Tencent's domestic game revenue grew 17%, marketing services revenue grew 22%; fintech and business services revenue grew 9%, with business services growth mainly coming from cloud services and AI-related demand. AI is also directly improving advertising recommendation efficiency.

Tencent's advantage lies in gaming, advertising, WeChat, and a large number of internal business scenarios. AI can first validate value in these scenarios and then export it to enterprises through cloud and agent products.

In overseas markets, cloud businesses are likewise bearing growth tasks.

The latest third-quarter financial reports show that Amazon's cloud business revenue grew 39.3%, Google Cloud grew 103%, and Microsoft's Azure cloud platform grew 48%.

They also emphasize leadership, but they have replaced market share with capability assessments.

AWS's official website uses "the world's most comprehensive and broadly adopted cloud"; Microsoft has long used Azure's growth rate to prove itself; four months ago, Google announced that in Gartner's first AI infrastructure report, its execution ability and completeness of vision were leading.

The difference lies in the customer structure.

According to Jefferies estimates, in the second quarter of 2026, OpenAI and Anthropic together accounted for about 45% of Google Cloud Platform revenue, 25% of Azure, and 6% of AWS. These major customers have the ability to evaluate models themselves, and procurement negotiations mainly revolve around unit computing power prices, delivery times, and cluster scale.

By contrast, procurement by large Chinese government and enterprise clients follows another process. Bidding requires scoring, decisions require basis, and projects require documentation. A market ranking in a third-party report can enter bid documents, briefing materials, and supplier evaluations.

The "number one" emphasized by each company therefore must be pitched to its target customers: Alibaba Cloud, facing large government and enterprise customers, emphasizes full-stack capabilities; Volcano Engine, facing developers and internet companies, emphasizes invocation volume; Huawei Cloud has more advantages in markets such as government affairs and finance; Baidu AI Cloud emphasizes years of accumulated AI public cloud projects; Tencent places more emphasis on internal scenarios and agent products.

After all, before selling tokens becomes a break-even business, cloud service revenue is the safety net for the giants' AI research and development spending.