June 2, Beijing. A list titled “2026 China AI Agent Pioneers” was released.
I went through this list three times. Not only because it includes AI agent cases from over a hundred companies, but also because a few places made me stop and linger for a long time: Hubei Bank uses AI for code security auditing, BOE uses AI for database operations and maintenance, Shougang uses AI for hot-rolled production decisions, and Muyuan uses AI to monitor pig herd health. From banks to pig farms, from highway inspections to satellite testing, spanning more than 20 industries—AI agents are no longer in the “pilot” stage; rather, they are running in companies’ core business processes.
But on the same day, another figure was equally striking: Zhipu AI announced that it would list on the STAR Market, raising 15 billion yuan while also losing 4.7 billion yuan.
These two events that happened on the same day made the landscape of China's AI in June 2026 clear: the application layer is booming, while the capital layer is still bleeding.
What the four tracks of the Top 100 ranking actually reveal
This list was jointly initiated by the Internet Society of China and the Artificial Intelligence Industry Development Alliance of China, and hosted by BCS 2026. Compared with previous editions, the biggest difference in the selection criteria is that “application value” and “security and controllability” each carry half the weight. Technical parameters are no longer the only yardstick; how long an agent has run in real business, whether it is secure, and whether it can be replicated—these are the key factors.
This list is divided into four parts and is well worth a look:
I noticed one detail: the industry application track has the broadest coverage—there are entries in all eight subfields: government affairs, energy, finance, manufacturing, healthcare, agriculture and animal husbandry, aerospace, and education. This is not a proof of concept of “what AI can do,” but rather that “AI has already generated quantifiable value in actual business operations.”
Shougang's hot rolling production AI agents are running on real production lines, BOE uses agents to manage database operations, and the Zhongshan Municipal Government Services Data Administration Bureau uses agents on the government extranet to perform secure operations—these are not PowerPoint projects.
If the theme for enterprise AI in 2024 was “can it be used,” and the theme in 2025 was “should it be used,” then by June 2026 the theme has shifted: how to use it safely, how to scale it, and how to keep iterating.
IDC pours cold water: In ERP selection, look at data first, then AI
Around the release of the ranking, IDC published the report "China AI-Enhanced Enterprise ERP Market Share, 2025".
Yonyou has taken the top position in the AI-ERP market. More than 60% of medium-sized manufacturing enterprises have incorporated AI capabilities into the core criteria for their ERP selection. The three major scenarios of finance, supply chain, and production scheduling have produced quantifiable return data.
An electronic components company in Shenzhen shortened its month-end closing time from five working days to half a day. Aier Eye Hospital deployed 17 digital employees, increasing financial processing efficiency by nearly 8 times. A hardware factory with annual revenue of 120 million yuan used AI demand forecasting to reduce raw material inventory turnover days from 68 days to 34 days, releasing 2 million yuan in working capital. After implementing AI scheduling, Shuangliang Group increased the accuracy of production plan execution by 50%.
These figures look encouraging. However, IDC said something in its report that many people may overlook: First assess how solid your own data foundation is, then determine the AI-ERP entry modules based on business pain points, and during selection, have the system run on real business data for validation.
If your source data management is unclear, no matter how powerful AI-ERP is, it cannot be implemented.
I think this is the one sentence in all the materials this week that is most worth an enterprise CIO reading three times. The Top 100 list says “AI has entered core processes,” while IDC says “whether it can enter depends first on your data foundation.” One talks about the ceiling, the other about the floor. Both are right.
The “Fire” of AI Implementation
· BCS Top 100: 20+ industries, full coverage across four major tracks
· BOE/Shougang/Muyuan and others have entered the production stage
· 17 digital employees at AIER Eye Hospital boosted efficiency eightfold
· Hardware factory inventory turnover halved, freeing up 2 million
· Huawei Cloud 8.5 million developers, 50,000+ partners
The “Ice” of AI Implementation
· IDC reminder: first look at the data foundation, then at AI
· Zhipu lost 4.7 billion yuan, with a market value of 630 billion yuan
· Large model companies are generally not profitable
· Enterprise data governance remains the biggest bottleneck
· More than 60% of manufacturing enterprises choose AI-ERP, but there is still a gap in POC validation
Zhipu IPO loses 4.7 billion yuan, what's wrong with the large-model sector?
On June 1, Zhipu AI issued an announcement on the Hong Kong Stock Exchange: the board has approved an A-share issuance plan and intends to list on the STAR Market of the Shanghai Stock Exchange. This company, which had just listed in Hong Kong in January, is seeking an A+H listing in less than half a year. The fundraising scale is 15 billion yuan.
But opening its prospectus reveals: revenue grew rapidly in the first half, yet it recorded a loss of 4.7 billion yuan, with a market value of about 630 billion yuan. This combination of “losing 4.7 billion and being worth 630 billion” is hard to imagine in traditional industries. However, the large-model track now follows a different valuation logic: it does not look at current profitability, but at the generational technology gap and ecological niche.
Zhipu is not the only company burning money. DeepSeek, Moonshot AI, MiniMax, and Baichuan—the “AI Six Tigers”—not a single one has truly achieved profitability yet. They are all betting on the same thing: as model capabilities continue to improve and economies of scale emerge, once inference costs drop to a certain critical point, their business models will become viable.
What does this mean for enterprise users? Two judgments:
First, in the short term, there is no need to worry too much about the issue of supplier bankruptcy. Zhipu's ability to list in Hong Kong and subsequently advance its listing on the STAR Market shows that the capital market remains optimistic about large models. A market value of 630 billion yuan corresponding to 15 billion yuan in financing provides strong credit endorsement.
Second, but you need to be prepared for model switching.The landscape of large models is still being reshuffled. This week, NVIDIA released Nemotron 3 Ultra, and Huawei Cloud ModelArts Next supports on-demand routing of 15 SOTA models. In the future, enterprise AI architectures should “not be bound to any single model.” Choose the platform, not the model—this is the more prudent strategy.
Three Pieces of Advice from the Head of Enterprise Digitalization
This week was packed with information, but there was actually only one theme: AI has fully entered enterprises' core processes, just not in the way many people imagine.
First: do not wait for a “perfect AI solution” to appear; instead, start by solving specific pain points.None of the successful cases on the Top 100 list involved “implementing a company-wide AI strategy.” They all started from specific scenarios—Hubei Bank began with code security auditing, Muyuan began with pig herd health monitoring, and Shougang began with a hot-rolled production line. Once they achieved success, they then expanded horizontally. This approach is far more reliable than “first formulating a three-year AI plan.”
Rule Two: Every dollar spent on data governance is more valuable than every dollar spent on AI models.IDC's reminder is not just a polite remark. Among the manufacturing companies I have worked with, whether the data foundation is solid can make a 5- to 10-fold difference in the effectiveness of deploying the same AI-ERP system. It is not that AI is not good enough, but that the data is not good enough. When your material codes are still inconsistent, BOM data still requires a lot of manual maintenance, and financial account mappings still have many exceptions—under such circumstances, deploying AI for production scheduling or AI forecasting will inevitably be much less effective.
Article 3: Pay attention to things like NVIDIA OpenShell and Huawei Cloud AgentSphere. Once AI agents begin operating in enterprises, security governance will become the most difficult issue. The point is not whether you “want to manage,” but whether you “are able to manage”—whether you can see in real time what an agent is doing, whether you can shut it down with one click, and whether you can set permissions by role. This week, NVIDIA and Huawei Cloud simultaneously launched secure runtime environments for agents. This is no coincidence; rather, the market is telling you through products: this issue has reached the point where it must be solved.
100+ BCS-selected enterprises 20+ industries covered 60%+ manufacturing enterprises AI included in ERP selection Zhipu half-year loss of 4.7 billion yuan (yuan)
The first full week of June 2026 made me feel that artificial intelligence has moved beyond the 'believe it or not' stage, and beyond the 'whether you want it is up to you' stage. It has now entered the stage of 'how to use it, how to manage it, and how to calculate the cost.' At this stage, what is being competed on is no longer technical capability, but organizational capability and data capability.
This is the moment when enterprise digital leaders truly take the stage.
