On May 20, Kingdee held an AI summit in Shenzhen, released an "AI-native product," and jointly launched a set of "AI management" research results with Peking University HSBC Business School.
The choice of this timing is quite interesting upon closer examination — the next day (May 21), the national AI agent evaluation standard seminar was submitted for review. Kingdee spoke out one day in advance, gaining a head start in the cognitive positioning of "AI management."
This matter itself is not news. What is worth noting is: the AI narrative of local ERP vendors is shifting from "feature stacking" to "management restructuring". What does this shift mean for the competitive landscape of the ERP market?
What are local manufacturers competing for behind the 48.7 billion market
In 2026, China's ERP market size is approximately 48.7 billion yuan, with local vendors holding a 68.4% share, an increase of 12.2 percentage points from last year. The cloud-native penetration rate has exceeded 51.2%, and products with deep AI integration account for over 65%.
The numbers look impressive, but when you break them down, the core of competition has already changed—
In the past, local vendors competed with SAP and Oracle for market share by relying on cost-effectiveness + localization. Now this logic is weakening because SAP and Oracle are also doing two major things:
- SAP Sapphire 2026 released the "Autonomous Enterprise", with 50+ Joule agents directly executing business processes, not just assisting.
- Oracle is also advancing a similar path, upgrading AI from a "helper" to an "executor."
This means that if local manufacturers still stay on the narrative of "our AI can help you automatically generate reports," the gap will widen again.
So Kingdee's current strategy—upgrading the narrative from "function" to "management"—is the right direction. But whether it can be implemented specifically depends on a few things.
Kingdee's "AI Management": Concept First, Implementation to Be Seen in Three Years
Based on public information, the core content released by Kingdee this time consists of two parts:
- Product Level: AI-native products (Kingdee AI Xinghan, Kingdee AI Xingkong, Kingdee Lingee, etc.), emphasizing "AI-native" rather than "AI overlay"
- Research Level: Jointly released "AI Management" research findings with Peking University HSBC Business School, elevating AI from a technical topic to a management topic
The cleverness of this approach lies in: it shifts the dimension of competition from "whose model is stronger" to "who better understands the AI transformation of enterprise management". SAP talks about "autonomous enterprise," while Kingdee talks about "AI management"—one leans toward a technological vision, the other toward a management methodology, with different audiences and discourse systems.
But there are several issues that need to be addressed—
First, what exactly does "AI-native product" mean? Does it refer to generating code from scratch using AI, or natively integrating AI capabilities at the product functionality level? There is currently no industry consensus on this concept. If Kingdee can provide a clear definition and comparable cases within this year, its first-mover advantage will be very significant.
Second, can the "AI Management" research methodology be productized? If PKU HSBC's research results only remain at the white paper level, their influence will be limited; if they can be turned into features within Kingdee's products such as management diagnostics, process optimization suggestions, and organizational design recommendations, the value will be much greater.
Third, how long is the time window? SAP's "autonomous enterprise" narrative has already been rolled out globally, and Yonyou and Inspur are also promoting their own AI products. Kingdee needs to deliver verifiable customer cases within 6-12 months, otherwise its cognitive advantage will be diluted.
What Yonyou and Inspur Are Doing
Expanding the view to the entire domestic ERP market, the differentiation in AI narratives among various players is becoming apparent—
| Manufacturer | AI core narrative | Current stage |
|---|---|---|
| Kingdee | "AI Management" + AI Native Products | Concept released, cases being accumulated |
| Yonyou | Enterprise Agent Platform (YonBIP AI) | Platformization stage, emphasizing ecology |
| Wave | Manufacturing AI + State-owned Cloud Scenarios | Deep cultivation of vertical industries |
| SAP (China) | Self-operated enterprise + Joule Agent | Global narrative, adapting to localization |
Yonyou's approach is different from Kingdee's. Yonyou emphasizes the "enterprise intelligent agent platform" — essentially packaging AI capabilities into a platform, allowing ISVs and customers to build their own intelligent agents on it. This line of thinking leans more toward ecosystem logic, representing a different business model from Kingdee's "I define the AI management methodology."
Inspur, on the other hand, is deeply cultivating manufacturing and state-owned enterprise scenarios, with its AI narrative being more specific: not "what AI can do," but "to what extent AI can help you in steelmaking, assembly, and warehousing." This vertical approach is highly effective in specific industries, but its ceiling is also evident.
Will the National AI Agent Standard Be a Variable?
The National AI Agent Evaluation Standards Seminar on May 21 is another noteworthy action this week.
This standard is jointly endorsed by three departments: the National Internet Information Office, the National Development and Reform Commission, and the Ministry of Industry and Information Technology. It covers four dimensions: value quantification, scenario list, security boundaries, and full lifecycle closure. Simply put, in the future, enterprises purchasing AI agents will have a national standard to refer to.
This is both an opportunity and a pressure for ERP vendors—
The opportunity lies in: with national standards backing, the decision threshold for enterprises to purchase AI will be lowered. Previously, CIOs had to bear the risk of "is this AI reliable or not," but now with standards in place, they can select and accept products based on these standards.
The pressure lies in: once the standard is implemented, it becomes an entry barrier. AI functions that do not meet the "value quantification" and "safety boundary" requirements in the standard may be excluded from government procurement and state-owned enterprise procurement.
Kingdee held a press conference one day before the standard was released, likely with the timing in mind—to first push out its own "AI management" framework, striving to align its framework with the standard by the time the standard is officially released, or even to influence the interpretation of the standard.
What is the actual impact on businesses
Translating these dynamics into the everyday language of enterprise CIOs and IT leaders, it roughly boils down to these matters—
First, when selecting ERP, the weight of AI capability evaluation will continue to rise. IDC data shows that 44% of CIOs have already made AI capability the primary criterion for ERP procurement. This proportion will continue to increase, especially among state-owned enterprises and large private companies.
Second, the gap between "AI-native products" and "AI-overlaid products" will become apparent in 18-24 months. The current difference may seem like just marketing jargon, but the underlying architectural differences will lead to divergence in subsequent customization costs, iteration speed, and data integration capabilities. When selecting a product, it is recommended to ask the vendor clearly: Is this AI feature added later, or was AI considered from the product design stage?
Third, after the national standard is implemented, there will be clearer acceptance criteria for the procurement of AI agents. CIOs should pay attention to the progress of this standard in advance and incorporate it into their AI procurement framework for the second half of 2026. Waiting until the standard is officially released before following up may delay the procurement pace.
At the end of the day, competition in ERP has never been about comparing feature lists, but about "who understands my business better." AI has made this question even harder—because now we also have to answer "who understands how my business is being transformed by AI."
