ByteDance's cloud division, Volcengine, has pulled off a strategic surprise in China's AI market, capturing nearly half of enterprise demand for AI models while Alibaba Cloud, the traditional cloud leader, watches its influence erode. In 2025, Volcengine commanded 49.5% of token call volume (a measure of how often companies use AI models) compared to Alibaba Cloud's 28%, according to IDC data. This shift reveals a fundamental tension in how companies are choosing their AI partners, and it mirrors a pattern ByteDance has executed before.

Why Is ByteDance Winning in AI Cloud When Alibaba Owns Traditional Cloud?

The answer lies in how the two companies approach the market differently. Alibaba built its cloud business by selling computing infrastructure first, then layering services on top. ByteDance took a different route: it bypassed the infrastructure battle and went straight for what companies actually want to use, which is access to powerful AI models through simple application programming interfaces (APIs). Think of it like the difference between selling someone a kitchen versus selling them prepared meals.

The Chinese AI cloud market was worth 56.7 billion yuan in 2025, split into two main categories. Infrastructure services, which sell raw computing power, account for about 69% of the market. Model-as-a-Service (MaaS), which packages pre-trained AI models into easy-to-use interfaces that companies pay for based on usage, accounts for about 31%. Alibaba dominates the infrastructure side, but Volcengine has captured the hearts and wallets of companies building AI applications.

When measured by overall revenue across the entire cloud stack, Alibaba Cloud leads with 38.1% market share, followed by Volcengine at 20.4%, according to Omdia. But the IDC report tells a different story: in the MaaS segment specifically, where the real growth is happening, Volcengine held 49.2% of token calls in the first half of 2026 without losing ground. Large model API calls in China's public cloud soared to 194.4 trillion tokens in 2025, a 16-fold increase year-over-year, and Volcengine captured nearly half of that explosive growth.

What Makes ByteDance's AI Models Stand Out?

The core reason companies are choosing Volcengine comes down to model quality and speed of innovation. ByteDance has released a series of competitive AI models that keep pace with global leaders. The company launched Seedance 2.0, which went viral during China's Spring Festival, followed by Seedream 5.0 for image generation and a voice-generation model that earned praise from Google DeepMind researchers. For AI application developers, using the latest and best models is a survival requirement; they need cutting-edge capabilities to retain users and stay competitive.

"Model capabilities are the key behind this. 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," said Cheng Ren, a post-training technician from a leading large-model manufacturer.

Cheng Ren, Post-Training Technician at Leading Large-Model Manufacturer

Alibaba's Tongyi Qianwen models, by contrast, have faced internal disruption. In summer 2024, Zhou Chang, the technical leader of Tongyi Qianwen, departed Alibaba and joined ByteDance's Seed team, bringing more than a dozen key team members with him. Zhou Chang had been instrumental in developing Alibaba's foundational models, including the M6 series, and his departure marked a turning point in the company's technical trajectory. The loss of continuity in leadership affected model iteration at a critical moment when the field was advancing rapidly.

How to Understand the Shift in China's AI Cloud Market

• Infrastructure vs. Intelligence: Alibaba built dominance by selling computing power and infrastructure services first, but ByteDance bypassed that layer and went directly to selling AI model access, which is what companies actually want to build with.
• Speed of Model Innovation: ByteDance released multiple competitive models in rapid succession, including Seedance 2.0, Seedream 5.0, and advanced voice generation, while Alibaba faced leadership departures that slowed its model development cycle.
• Market Segment Focus: While Alibaba leads in overall cloud revenue, ByteDance captured the fastest-growing segment, MaaS, where token call volume grew 16-fold year-over-year and Volcengine holds nearly 50% of the market.
• Enterprise Betting Patterns: Companies are building their application ecosystems on Volcengine's models rather than Alibaba's infrastructure, suggesting a shift in how enterprises evaluate cloud vendors in the AI era.

This dynamic echoes a pattern ByteDance has executed before. A decade ago, Alibaba defined e-commerce with product categories and logistics infrastructure, but ByteDance captured user attention with Douyin (its short-video platform), forcing merchants to open stores there. Now, in the AI cloud market, Alibaba defined the space with full-stack capabilities and government-enterprise relationships, but Volcengine captured customers' minds with superior models. The logic is identical: ByteDance uses new paradigm traffic and capabilities to penetrate markets defined by older business models.

The implications extend beyond market share numbers. If this trend continues, ByteDance could reshape the power structure of China's internet industry over the next decade. Alibaba's management has acknowledged the challenge internally, with insiders describing Volcengine's success as "the most painful thing for Alibaba this year." Despite Alibaba Cloud's stated goal of capturing 80% of incremental AI cloud market growth in 2026, the first half of the year showed Volcengine maintaining its lead without dilution.

For enterprises evaluating cloud providers, the message is clear: in the AI era, model quality and innovation speed matter more than traditional infrastructure dominance. Companies racing to build AI applications are voting with their token consumption, and they are choosing ByteDance.

" }