The fifth battle of Chinese cloud vendors has begun!

Since the starting gun fired in 2007, in the blink of an eye, China's cloud computing has been sprinting for nineteen years.

In the blink of an eye, years have passed, and clouds have filled the sky.

Nowadays, the bugle of the fifth upheaval has sounded once again. Over the past 19 years, from hardware virtualization, platformization, and cloud-native, to AI integration, to put it without exaggeration, cloud computing is the core foundation for China's Internet to build a domain-wide unified computing power base and finalize the underlying digital technology architecture.

At the same time, cloud computing has also reshaped the form of existence and the underlying logic of value circulation of China's Internet digital resources, promoting China's Internet to fully transition from the consumer traffic era to the industrial digitalization era.

From cost savings to efficiency improvements, from application innovation to business empowerment, in today's world where large models are developing like wildfire, the identity of cloud computing has once again achieved a value leap—moving towards value creation. From IaaS to PaaS, from XaaS to MaaS/AIaaS, in 2026 Chinese cloud computing vendors collectively and tacitly found industry consensus once again—Agent Infra has become the core strategic positioning.

Since Q2, the bugles for the "Fifth Campaign" among cloud vendors have been sounding one after another. On May 13, Robin Li, founder of Baidu, Inc. / Baidu (百度), proposed that DAA should become the metric of the AI era, rather than DAU. As the second-generation entry point, the value ceiling of Intelligent Agents is far higher than that of chatbots. At the same time, he also proposed three levels of "self-evolution" in the AI era, and the self-evolution of Agents is one of the key levels.

A week later, at the just-concluded 2026 Alibaba Cloud (阿里云) Summit, Liu Weiguang, Senior Vice President of Alibaba Cloud Intelligence Group (阿里云智能集团) and President of the Public Cloud Business Unit, stated even more bluntly: "Cloud infrastructure is an important technical cornerstone of the Agentic era." Only brand-new cloud infrastructure can meet the diverse needs of Agent operation, such as stability, security, and timely scheduling. And its newly released "Qwen Cloud" (千问云), It is even called "a brand new service mode born for Agent."

In fact, in the following June, cloud vendors like Tencent Cloud (腾讯云) and Volcano Engine (火山引擎) could no longer hold back, Tencent Cloud's (腾讯云) AI Industry Application Conference and Volcano's (腾讯云) AI Engine's (火山引擎) Volcano Engine FORCE Conference (火山引擎FORCE原动力大会) are about to be held. Agent has become another growth point for AI, and the transformation of infrastructure services closely related to it has become a new "story".

From AI Infra to "Agent New Infrastructure" (Agent新基建)

In the past three years, the core of cloud computing has been AI Native Cloud (AI Native Cloud), essentially "computing power optimization", It is a computing power cluster centered around large model training and inference, providing high concurrency, high throughput, and low latency. The core metrics are cluster scale, computing power density, and network bandwidth.

However, as Agents penetrate into more scenarios, traditional AI infrastructure can no longer meet the native needs of Agents. The industry consensus is that the cloud must shift from "serving models" to "serving Agents." "Computing power utilization rate" is no longer the sole core metric; "Agent task success rate, execution efficiency, and governance controllability" have become the measuring factors. EqualOcean learned at this Alibaba Cloud (阿里云) summit that currently, Alibaba Cloud (阿里云) breaks this down into two layers.

First is the AI native cloud, continuing to deeply cultivate computing power support for model training and inference. Second is the agent native cloud, building infrastructure specifically for the orchestration, operation, and governance of agents.

Correspondingly, Baidu (百度) previously took Agent Infra as the core and proposed the "Chip-Cloud-Model-Device" full-stack architecture, upgrading MaaS to Token Factory. The core is to enable every Token to be efficiently converted into executable intelligent actions.

The essence of this shift is a fundamental change in the target of cloud computing services: shifting from services for "software written for humans" to services for "agents that make autonomous decisions and execute automatically." Traditional clouds are oriented towards deterministic tasks, with long-term resource occupation and stable loads. However, Agent tasks have four major characteristics: short lifecycle, irregular bursts, dynamic dependencies, and task-level security—an agent might initiate a task in one second and be destroyed in the next, or it might run continuously 7×24 hours; it also needs to frequently call databases, browsers, and third-party tools. This requires cloud infrastructure to shift from "resource scheduling" to "task scheduling".

At the 2026 Baidu Create conference, Baidu (百度) founder Robin Li stated bluntly: "Tokens don't necessarily represent the endgame; DAA (Daily Active Agents) is the new metric for the AI era." In the past, the industry competed over who burned more Tokens and who had larger clusters; now, the competition has shifted to who can support more Agents to work stably and deliver results. Alibaba Cloud (阿里云) also offered a similar judgment: "We are moving, for the first time, from the large-scale management of computing power to the large-scale management of intelligence." The consensus between these two giants announces a thorough reconstruction of the underlying logic of cloud computing.

The "acclimatization issues" of traditional infrastructure

The shift in AI Infra is no accident; there is a fundamental mismatch between traditional cloud computing architecture and the native requirements of Agents. Therefore, the shift in AI Infra is a necessary requirement for Agents to move into scenarios.

On one hand, traditional resource scheduling has become "ineffective" in the "computing power era". Agent workloads are completely different from traditional AI tasks and internet tasks. They have a short lifecycle; most Agent tasks take seconds to minutes and are destroyed immediately after use. They are sporadic and irregular, with traffic potentially exploding ten thousand times in an instant or remaining dormant for long periods. Moreover, they have strong state dependency, requiring continuous memory of context and tool call history. They involve multimodal interactions, frequently calling diverse tools such as text, images, video, and databases. Traditional clouds are designed based on "long-cycle deployment and stable loads," making them unable to adapt to this "pulsed, stateful" load. For example, traditional containers take minutes to start, which cannot support the second-level start and stop of Agents. Resources are billed based on long-term instances, which cannot match the cost requirements of Agents' "short-term high load, long-term dormancy." Alibaba Cloud (阿里云) research found that when enterprises build their own Agent platforms, just the container cost alone far exceeds expectations.

On the other hand, there is the contradiction between cost and efficiency. The high cost of large model inference is further exacerbated by the multi-round calls and repeated context calculations of Agents. Data from Baidu (百度) shows that in traditional MaaS services, about 30% of Tokens are used for repeated calculations, resulting in low inference efficiency. Alibaba Cloud (阿里云) also mentioned that when the KVCache hit rate is less than 70%, the memory bottleneck in inference leads to a plummet in efficiency. At the same time, the implementation of enterprise Agents faces the reality that "95% of tasks are repetitive labor, and 5% are core decisions." Traditional infrastructure cannot reuse historical calculation results, leading to "calculating from scratch for every call," and costs remain high. Shen Dou, Executive Vice President of Baidu (百度), pointed out: "In the Agent era, cost is not computing power cost, but Token efficiency cost."

In addition, security governance is also one of the reasons that cannot be bypassed, Agents can autonomously access core enterprise data, invoke business systems, and execute operations. The traditional cloud security system (account permissions, network isolation) follows the logic of "humans using software" and cannot adapt to the needs of Agent identity authentication, refined permission control, behavior auditing, and data leakage protection. For example, if an Agent accidentally deletes a database or leaks customer data, the traditional security system cannot trace accountability or intercept it in real time. Furthermore, when multiple Agents collaborate, issues such as memory sharing, permission isolation, and task conflicts are completely unaddressed by traditional governance tools. Alibaba Cloud (阿里云) has summarized the six core challenges for enterprise Agent implementation, with security and governance accounting for three of them. This is also the core reason why most enterprises "dare to do a Demo but dare not go into mass production."

Agents will reconstruct industries, and new infrastructure is the entry ticket. Robin Li previously predicted at the Baidu (百度) AI Developer Conference that in the future, the global number of DAAs will exceed 10 billion, and every position and every scenario will have multiple Agents handling tasks. Alibaba Cloud (阿里云) estimates that in the next 2-3 years, Agents will witness explosive growth, with enterprise workflows shifting comprehensively from "human-centric" to "Agent-centric."

Data from PwC (普华永道) shows that 79% of US companies are already using Agents in their business operations, and 88% plan to increase investment. Gartner predicts that by 2028, 33% of enterprise software will natively integrate Agent capabilities. Facing this definite trend, cloud vendors must take the lead in laying out Agent-Native infrastructure; otherwise, they will lose their ticket to the next round of industrial competition.

The pivoting paths of Alibaba Cloud (阿里云) and Baidu (百度) are highly similar, and AI full-stack capabilities based on "chip, cloud, model, and device" have become an important direction.

Among them, the most critical hardware is the chip. The requirements of agents for chips are "high inference performance, low latency, high concurrency, and low cost," which traditional GPUs cannot meet, so both giants are focusing on developing their own chips.

Alibaba Cloud (阿里云) launched the Zhenwu M890 (真武M890) training and inference integrated AI chip, paired with the ICN Switch 1.0 networking chip, and mounted on the Panjiu AL128 Supernode Server (磐久AL128超节点服务器).

It is understood that this chip is optimized for Agent inference, supporting 800Gbps high-speed networks. A single cluster can support a scale of 100,000 cards, with a linear expansion efficiency of over 96% for 10,000 cards, focusing on addressing the computing power demands of high-concurrency Agent inference and short-term high loads. Correspondingly, Baidu's self-developed Kunlun Chip (昆仑芯) has iterated to the P800, continuing to deliver 10,000-card-level clusters. The Tianchi 256-card Supernode (天池256卡超节点) will be launched in June, improving inference efficiency by 50%.

Kunlun Chip (昆仑芯) is deeply adapted to the ERNIE large model (文心大模型), while also supporting mainstream models such as DeepSeek and Zhipu GLM / GLM (智谱GLM); the core is to improve Token production efficiency and reduce Agent invocation costs. In addition, the architecture layer, product layer, and overall ecosystem capabilities will all undergo adjustments and changes as Agents enter more scenarios and workflows.

It is worth noting that the shift in Agent infrastructure by cloud vendors is not only a technological change, but also an industrial transformation. One manifestation is "the reduction of costs". Although the reduction of costs is not yet a direct result at the moment.

Xu Qing (许青), President of the Terminal Intelligence Computing Business Unit at Alibaba Cloud Intelligence Group (阿里云智能集团), stated in a media interview while introducing JVSClaw: "The launch of JVSClaw has brought about some demands from B-side clients, and the technical team is able to abstract the capabilities required for Agent Infra from these client needs. Therefore, even if it means doing this at a 'loss' for now, it is still worth it."

Besides cost, controllable security governance has become a major advantage. Capabilities such as Agent identity authentication, refined permission management, behavior auditing, and data isolation address core enterprise concerns. Alibaba Cloud (阿里云) Agent Security Center can trace every Agent operation, and Baidu (百度) AI Security Guardrails can intercept abnormal behavior in real time.

However, as Agent becomes the next "explosive point" captured by the market and the industry, the trend of industry competition and "accelerated reshuffling" has become apparent. Cloud vendors are shifting from selling computing power to selling "intelligence services"—Agent orchestration, memory management, task execution, etc., will become new revenue growth points.

Giants with full-stack capabilities, such as Alibaba Group (阿里巴巴集团) and Baidu (百度), are seizing the first-mover advantage; if small and medium-sized cloud vendors cannot keep up quickly, they will be marginalized. Vertical domain solution providers will rise, focusing on industry Agent scenarios such as finance, manufacturing, and healthcare.

The shift towards Agent-based cloud computing is essentially a triple resonance of technology, industry, and capital. Robin Li (李彦宏) said: "In the future, what matters is not computing power, but intelligence." Meanwhile, Alibaba Cloud (阿里云) believes: "Tokens must become intelligence, and intelligence must become action." This transformation has only just begun.

Alibaba Cloud's (阿里云) Agentic Cloud and Baidu's (百度) Agent Infra are just the starting point, not the destination. In the future, the core competitiveness of cloud vendors will no longer be cluster scale or computing power density, but the intellectual operational capability to support the efficient, secure, and stable operation of massive numbers of agents. For enterprises, embracing the Agent era means embracing new productivity.

For cloud vendors, winning the Agent infrastructure war is the only way to seize the commanding heights of the industry for the next decade. From computing power to intelligence, the second half of cloud computing has just begun.