Cloud vendors face an AI dilemma: Baidu goes left, Alibaba goes right.

The competitive landscape of China's AI cloud market has never been more uncertain in the second half of 2025.

At Beijing Capital Airport, travelers can see advertisements from three cloud vendors with varying postures: Alibaba Cloud claims 'leading AI cloud market share, surpassing the combined total of the 2nd to 4th place,' Volcengine emphasizes 'occupying 46% of China's public cloud large model market share,' and Baidu Intelligent Cloud boasts 'leading China's AI public cloud market share for six consecutive years.'

Each company selects statistics that favor its position. Market research firms IDC, Omdia, and Frost & Sullivan released at least seven reports in 2025, depicting the emerging market from different dimensions.

Baidu and Alibaba, the most representative players, embody two distinct strategic approaches. Baidu views AI cloud as a pillar for core transformation, seeking breakthroughs through vertical technology integration. Alibaba, on the other hand, leverages its full-stack ecosystem and international layout (global layout) to build broader industry influence.

Behind the battle for 'number one,' a profound shift is underway: the AI cloud market is transitioning from technology-driven to commercial applications, from model training to inference services, with a price war quietly igniting and token consumption emerging as a new competitive metric.

The divergent paths of the two companies reflect common challenges faced by China's AI cloud industry: balancing short-term market share with long-term profitability and transforming technological capabilities into sustainable commercial value.

01

From Computing Power to Tokens: AI Cloud Reaches a Turning Point

The AI cloud market has encountered two inflection points reshaping the landscape in 2025.

The first inflection point occurred in July 2024 when Doubao's large model achieved technological cost reduction, lowering inference prices from 'per-cent' to 'per-thousandth' pricing and reducing costs by up to 99.3%, enabling widespread AI adoption.

The second inflection point came in February 2025 when the market grew by 60%, coinciding with the popularity of the inference model DeepSeek-R1, marking China's AI market's official entry into the 'inference era.'

A recent IDC report revealed that external calls for large models on China's public clouds reached 536.7 trillion tokens in the first half of 2025, nearly quadrupling from 114 trillion tokens in 2024.

This data underscores the rapid growth and market transition of China's AI industry. IDC emphasized, 'Token consumption has become a barometer for the AI industry. Each token represents a genuine model output, whether text generation, image recognition, or voice interaction, reflecting the tangible integration of AI with business scenarios.'

From a daily perspective, token consumption growth is even more astonishing—rising from 100 billion daily in early 2024 to over 30 trillion by June 2025, a 300-fold increase in just one and a half years.

Behind the explosive market growth lies a shift in enterprise demand from focusing on model training accuracy to pursuing 'cost efficiency and sustainability' in inference. The industry's core contradiction is shifting: enterprises are moving from 'owning models' to 'using models effectively,' with tokens becoming a new metric for AI cloud value, directly reflecting model efficiency in real-world scenarios.

For cloud vendors, as traditional IaaS revenue growth stalls and MaaS token calls become a growth engine, the industry exhibits three distinct features: application scenarios shift from training to inference, competition focuses from computing power to ecosystems, and value propositions transition from resource provision to capability output.

'The industry is undergoing a shift from selling water to helping people pan for gold,' a cloud computing industry analyst remarked.

This reality poses complex challenges for cloud vendors. Early on, cloud vendors primarily provided computing resources, akin to selling water to gold prospectors, with core competitiveness rooted in scale effects and resource management. Now, leading vendors offer complete AI capability stacks, requiring comprehensive abilities in model development, ecosystem operations, and industry understanding to help clients create stable business value directly.

02

Baidu Intelligent Cloud: Pioneering AI Cloud with Hidden Concerns

From its development trajectory, Baidu Intelligent Cloud can be seen as an early explorer in China's AI cloud market.

As early as 2016, Baidu began building its 'cloud + AI' strategy, at least three years ahead of industry consensus. Baidu founder Robin Li repeatedly emphasized in public: 'Baidu has always been an AI company; cloud is merely an outlet for AI capabilities.'

However, its uniqueness stems not simply from its early entry but from two critical nodes.

First, its technology-first vertical integration strategy forms the core of Baidu's AI cloud approach. Li's emphasized 'four-layer AI architecture'—spanning chips, frameworks, models, and applications—creates an end-to-end closed-loop system.

This architecture has proven effective in cost control: Wenxin's large model achieved an 80% price reduction while enhancing performance. For instance, the input cost of Wenxin 4.5 Turbo dropped to 0.8 yuan per million tokens, attempting to rapidly capture market share through price advantages.

On the other hand, Baidu Intelligent Cloud President Shen Dou pointed out that Qianfan's approach of providing the lowest inference costs and highest concurrent stability makes it an entry point for enterprises to adopt AI with low barriers—a crucial growth source for Baidu AI. The Qianfan platform has integrated over 100 mainstream models, with enterprises and developers fine-tuning 33,000 models and developing 770,000 enterprise applications through the platform.

To date, the platform has served over 460,000 enterprise clients, with the number of agents developed on it exceeding 1.3 million. Daily calls for tool components, represented by Baidu's exclusive 'AI Search,' have surpassed 10 million.

As widely known, Baidu was once synonymous with search engines. With traditional search growth plateauing, AI has become its most critical transformation direction. The Q1 2025 earnings report showed a 42% YoY increase in AI cloud business within Baidu's core revenue, contributing 26% of core revenue.

However, behind a series of growth figures, Baidu AI cloud still faces various structural hidden concerns.

For instance, its customer base is overly concentrated. Despite leading in project wins—48 projects and 510 million yuan in the first half of 2025—a significant portion comes from government agencies and state-owned enterprises, with limited coverage of small and medium-sized enterprises. This reflects a mismatch between its technological offerings and market demand, indicating insufficient penetration in mainstream commercial markets.

Coupled with the price war eroding profit margins, Baidu was forced to join price reductions to compete with Volcengine and others. However, as ACG President Shen Dou admitted, 'domestic price wars result in significantly lower revenue than abroad.' In the era of 'per-thousandth' token pricing, scale growth has not synchronize (synchronized) brought profit improvements, creating a paradox of growth amidst anxiety.

Additionally, Baidu's heavy reliance on API interface models, compared to Alibaba Cloud's open-source ecosystem, has led to lag (lagging) developer ecosystem construction due to technological closure (closed nature). Some analysts noted that partner enterprises switched to other platforms due to high technical adaptation costs, weakening their participation in industrial collaboration.

In industry insiders' views, Baidu's challenges may extend beyond direct competition to addressing the rise of the token economy.

IDC pointed out that with the emergence of multimodal technologies and intelligent agent applications, token consumption per task will grow exponentially, and the inflection point for scalable AI applications is approaching. Only those providing the most cost-effective and stable model inference services can attract the most developers and dominate the token economy of the AI industry.

03

A Battle Between Baidu and Alibaba Is Inevitable

Unlike Baidu's focus on core AI cloud business, Alibaba Cloud exhibits a more pronounced diversification (diversified) strategic layout.

The open-source nature of its technology ecosystem is a defining feature of Alibaba Cloud, with its Qwen series models open-sourcing over 300 AI models, surpassing 600 million downloads, and spawning over 170,000 derivative models. Alibaba Cloud's revenue grew 18% YoY in Q4, with public cloud services outpacing overall cloud revenue growth; AI-related product revenue maintained triple-digit growth for seven consecutive quarters.

Meanwhile, Alibaba's AI strategy is integrated into its vast ecosystem, encompassing China's consumer market (Taobao, Tmall), international retail platforms (AliExpress, Lazada), wholesale trade, logistics (Cainiao), local services, and digital media.

In 2025, Alibaba Cloud emerged from a three-year slump. In the first half, its revenue reached 63.5 billion yuan, up 21.8% YoY, matching the growth rate of the top three international cloud vendors. With AI enhancement, Goldman Sachs estimated in March that Alibaba's AI infrastructure investment in FY2025 would generate approximately 30 billion yuan in annualized AI revenue.

Another advantage of Alibaba Cloud lies in its international layout (global layout). Recently, Alibaba Cloud announced the official launch of its second data center in Dubai, marking another new facility globally this year. Its global footprint now spans 29 regions and 92 availability zones.

In China's AI cloud market competition, the rivalry between Baidu and Alibaba has evolved from traditional cloud computing services to a clash of two AI strategic paths: Baidu's 'vertical integration, AI-native' deep focus versus Alibaba's 'full-stack ecosystem, industry empowerment' broad expansion.

The core of the competition remains the struggle for dominance in the AI cloud market. Baidu Intelligent Cloud leverages its early AI layout (layout) to solidify its positioning as an AI-native cloud service provider. Alibaba Cloud, relying on its vast cloud computing foundation and ecosystem, emphasizes its comprehensive capabilities as an integrated cloud service provider in the AI era, proposing the vision of 'AI cloud as the next-generation computer.'

The differences in their technological paths do not prevent them from vying for the same customer base, such as in the government and enterprise market. In the first half of 2025, Baidu Intelligent Cloud led in project wins and contract value in China's large model bidding market, but this reflects its heavy reliance on central and state-owned enterprise clients. Alibaba Cloud, leveraging its multi-industry client base accumulated over the years, seeks penetration in a broader enterprise market.

The 'inevitable battle' between Baidu and Alibaba is essentially a contest over the path to maturity for China's AI industry.

In the short term, Alibaba will likely maintain its revenue lead through full-stack capabilities and ecosystem advantages, but Baidu retains irreplaceable value in AI-native experiences and vertical industry depth.

The key to medium-term competition lies in the ability to implement 'industry AI transformation.' Whoever can better understand industry pain points and provide genuine problem-solving solutions, rather than merely offering technological tools, will win the long-term trust of enterprise clients.

In the long run, the competitive boundaries of the two companies may gradually clarify, forming a differentiated landscape of 'vertical depth' and 'ecosystem breadth.' The Chinese market can accommodate multiple business models; the key is finding a sustainable growth path.

The current AI cloud competition has entered a strategic stalemate, with players consolidating strengths and addressing weaknesses. The shared challenge is proving that AI technology can translate into widespread commercial success and demonstrating that full-stack capabilities can produce differentiated AI value. The future belongs to those who can both grasp technological trends and deeply understand industrial essence.