The cloud computing market has reached its 'pivotal moment,' akin to the transformative impact of the iPhone's introduction.

Recently, CNR reported that AI large models are entering a critical phase of implementation, and the Chinese cloud computing market is undergoing a significant industrial restructuring.

Leading domestic providers are transitioning from competing on hardware unit prices to focusing on token output efficiency in the realm of 'AI + Cloud Computing.' There are new strategic moves in large-scale localized computing power deployment and Agentic Infra (agent infrastructure).

In this silent war, leading cloud computing providers Aliyun, Tencent Cloud, Huawei Cloud, and Baidu Intelligent Cloud have each unveiled their strategic advantages.

Aliyun's Strengths and Vulnerabilities

Aliyun stands at a crucial juncture but remains the undisputed leader in the Chinese cloud computing market.

Aliyun has developed full-stack self-research capabilities in both hardware and software. It is the only cloud computing company in China with vertical integration capabilities, boasting its self-developed 'Apsara' operating system and Lingjun AI computing cluster, which support large-scale deployments of '100,000 computing cards' while integrating with 800Gbps low-latency networks. This makes it the preferred training service provider for trillion-parameter models.

At the chip level, Ali's T-Head Semiconductor has achieved large-scale production and industrial application of AI computing chips, with over 60% of computing power utilized by external customers.

Aliyun's open-source ecosystem attracts developers and indirectly controls traffic entry points. Through its 'Tongyi Qianwen Qwen' series of products, Aliyun has established the world's largest open-source model training base.

According to U.S. industry tracker 'Interconnects AI' in March this year, Tongyi Qianwen boasts over 200,000 derivative models, ranking first globally in open-source ecosystem scale. This vast open-source ecosystem generates diverse customized AI training demands, providing a long-term stable growth avenue for Aliyun.

Aliyun's global layout, combined with e-commerce consumption scenarios, creates a closed business loop. In June this year, Aliyun announced that its services now cover 32 regions and 105 availability zones globally. According to Gartner, Aliyun is now the largest cloud service provider in China and the first in Asia.

Internal consumption chains like Taobao and Ele.me are also gradually being integrated by Aliyun. This ability to embed large models into real-life scenarios is a competitive advantage that is difficult for rivals to replicate.

Despite its scale advantages, Aliyun faces two structural weaknesses.

Multiple strategic adjustments have shaken market confidence. Over the past few years, Aliyun has undergone several management changes, business structure splits, and strategic adjustments, leading to market uncertainty and causing some external customers to adopt a wait-and-see attitude.

In terms of government and enterprise customers, Aliyun needs to enhance its security capabilities. Although its e-government cloud service has passed regulatory network security Level 4 qualification review and can undertake non-sensitive government business, it has yet to meet the standards for bidding on projects involving sensitive departments and state-owned core systems.

Huawei Cloud's 'Authentic Approach' and 'Public Sector Stronghold'

Among the four, Huawei Cloud has fully built a localized hardware and software ecosystem, creating a competitive moat for information technology application innovation, making it the most 'authentic' player.

A fully localized closed-loop ecosystem ensures a stable base of government and enterprise users. From Ascend neural network processing units to Kunpeng central processors, Huawei Cloud has constructed a fully localized closed-loop ecosystem. For users in government, finance, and other sectors with stringent compliance needs, Huawei Cloud is one of the few providers that can meet their requirements.

In November last year, IDC reported that over 65 large central state-owned enterprises utilize Huawei Cloud, covering more than 70% of the market. In public customer sectors such as government clouds, finance, and central state-owned enterprises, Huawei Cloud has consistently ranked first in market share for multiple years.

Leading the introduction of agent infrastructure to compete for the definition rights of AI infrastructure. Huawei Cloud has transcended the traditional cloud computing service provider role and proposed a new technical system called the Token Factory.

In June this year, Huawei released four new underlying products, including the AICS Lingqu Intelligent Computing Cluster and AMS Memory Storage, attempting to seize the definition rights of infrastructure in the agent era. Huawei Cloud CEO Zhou Yuefeng emphasized at the launch event: 'Huawei Cloud's core goal is not just to pursue model parameters and token scale but to transform every token into real productivity and create a fertile ground for enterprise-level AI innovation.'

The Pangu large model focuses on vertical industries and has strong scenario penetration. Huawei Cloud digitizes advanced experiences in mining, healthcare, automotive manufacturing, and other fields through its Pangu series of industry large models, enabling small and medium-sized enterprise customers to quickly implement vertical scenarios without building their own computing power and model systems.

In the new era, Huawei Cloud also confronts new challenges.

Competition from telecom operators puts pressure on Huawei Cloud's public business. Telecom operators like China Mobile leverage their advantages in bandwidth and government and enterprise dedicated lines to attract customers at low costs, diverting some of Huawei Cloud's general-purpose computing orders for small and medium-sized customers.

Customer concentration in the government and enterprise sector leads to insufficient concurrent service capabilities for internet businesses. Compared to Tencent Cloud, Huawei Cloud lacks ecological cooperation with national-level applications like WeChat. There is also a gap in maturity for high-instantaneous concurrent businesses like live streaming, which may not meet the procurement needs of large internet companies.

Tencent Cloud's 'Distinctive Features' and 'Follow-the-Leader Approach'

Tencent Cloud is a unique player among the leading cloud providers, leveraging the billion-user social ecosystems of WeChat and QQ to gain advantages in C-end sectors like real-time audio and video and game globalization.

Long-term leadership in audio and video solutions with an exclusive ecological barrier. According to IDC's 'China Video Cloud Market Track,' Tencent Cloud has ranked first in China's audio and video market share for eleven consecutive years since 2018. Its social ecosystem serves as an exclusive barrier, with leading ecological fission capabilities in the industry.

NPO technology is the first to be implemented, with the potential to significantly reduce inference costs. NPO (Near-Package Optics) is a cutting-edge hardware technology in the cloud computing industry that shortens the optoelectronic conversion link and reduces hardware losses, potentially significantly lowering the comprehensive inference costs of large models, with industry estimates of up to a 40% reduction.

At the 2026 World Artificial Intelligence Conference in July this year, Tencent Cloud announced that it would deploy China's first large-scale NPO super node, expected to be launched and promoted in the fourth quarter of this year.

TokenHub integrates with WeChat, lowering the barrier for enterprise agents. Recently, Tencent Cloud upgraded its MaaS platform to the TokenHub platform, which, after integrating with WeChat, connects to native functions of Mini Programs and Enterprise WeChat, further lowering the barrier for using enterprise agents.

From a research and development strength and business structure perspective, Tencent Cloud's weaknesses cannot be overlooked.

Limited application scale of self-developed chips, and still in a following position for large model training. Tencent's self-developed Zixiao computing power chip is currently only piloted internally on a small scale, with external chips primarily relied upon for computing power production in external orders. Due to a lack of deep hardware-software collaboration capabilities, Tencent Cloud lacks competitiveness in trillion-parameter training projects.

Insufficient experience in large-scale projects for government and enterprise customers. Currently, Tencent Cloud adopts an 'ecological joint delivery' business model for large-scale public projects of government and enterprise customers, requiring the introduction of other partners. Compared to Huawei Cloud's independent delivery capabilities, Tencent Cloud's comprehensive capabilities in this type of business need improvement.

Baidu Cloud's 'All-In' Commitment and 'Technical Specialization'

While other cloud computing providers still treat artificial intelligence as a value-added service, Baidu Intelligent Cloud has already integrated AI into its core DNA.

Established the Baige heterogeneous scheduling platform, leading in multi-chip mixed training efficiency. Facing supply chain risks from single-chip vendors, Baidu Intelligent Cloud built the Baige heterogeneous scheduling platform, making its multi-chip mixed training efficiency industry-leading.

Its Baige computing power scheduling system can be compatible with various computing cards, including NVIDIA and Baidu Kunlun, achieving 95% efficiency in mixed training at the 10,000-card level. Enterprises like Postal Savings Bank of China and Changan Automobile have already adopted this training solution for their mixed computing power clusters.

Paired with knowledge graphs to facilitate enterprise privatization of data transformation. Baidu Intelligent Cloud, leveraging its Wenxin large model, has introduced a complete Retrieval-Augmented Generation (RAG) technology and a knowledge engineering toolchain, making it easier for enterprises to transform their privatized data.

Currently, Baidu Intelligent Cloud's data intelligence business has served 65% of central state-owned enterprises, over 800 financial institutions in China, and the top 15 automotive OEMs, with deep business accumulation.

Long-term accumulation in intelligent driving business, with embodied intelligence building an early advantage. Relying on Baidu's long-term investment in the Apollo intelligent driving platform, Baidu Intelligent Cloud can provide AI road training covering over one million kilometers daily. The company has also provided dedicated computing power to the Humanoid Robot Innovation Center, giving it an early advantage in the frontier field of humanoid intelligence.

From a scale and ecological perspective, Baidu Intelligent Cloud has relatively obvious weaknesses.

Relatively lagging in the construction of ultra-large computing power clusters and global layout. Compared to Aliyun and Huawei Cloud, Baidu Intelligent Cloud has a certain gap in training scale at the 100,000-computing-card level. Additionally, its overseas business layout started relatively late, with fewer procurement orders from foreign customers.

Insufficient presence in open-source ecosystems and developer communities. While Aliyun's Tongyi Qianwen has over 200,000 derivative models and ranks first globally in open-source large models, Baidu Intelligent Cloud's Wenxin large model only announced its open-source status in May last year, with a relatively slower progress in building its open-source ecosystem.

Final Showdown: Who Will Dominate the AI Era?

When comparing the four providers, the future development trend of the AI cloud computing market becomes clear.

Core competitive metrics have shifted, with unit token cost determining success or failure. As the cloud computing industry enters its next phase, the focus has shifted from comparing the number of computing cards to who can produce tokens at lower costs and higher efficiency. If cloud computing providers do not achieve breakthroughs in inference costs and system-level scheduling, they will be hindered by high computing power costs and fail to thrive.

The ecological locking effect is gradually strengthening, and single-platform models are becoming ineffective. The different technical routes adopted by the four companies indicate that there is no absolute winner in the cloud computing market. Future competition will be a battle of ecological and segmented scenario functions, making it difficult for a single platform to achieve optimal status in all dimensions.

The rise of natural language has shifted the focus of cloud computing industry operations. As natural language interaction becomes the mainstream in the AI era, the core competitiveness of cloud computing providers has shifted from 'managing servers' to 'operating digital employees.' Providers that can offer better agent services will control the next generation of traffic entry points.

A multi-polar competition landscape will persist long-term, with no single provider winning everything. Government customers prefer Huawei, e-commerce and retail suit Aliyun, audio and video businesses lean toward Tencent, and enterprise data and autonomous driving opt for Baidu. The four major providers each have their own advantage areas. Their collective efforts will significantly lower the cost of intelligent computing power across society, and the era of AI 'big smelting' may be upon us.

Only time will tell how the future of the cloud computing industry unfolds.