Hello! I'm your financial analysis assistant. This in-depth article about the Ministry of Industry and Information Technology's "Special Action for Cultivating Artificial Intelligence Application Service Providers" contains a wealth of information and hits right on the most critical issues in China's AI industry at the moment.
To help you easily understand the business logic behind this substantial policy, I've summarized the key points in one sentence and then broken it down into five key dimensions, explaining them in plain language.
📌 Summary of Key Points: From "Competing on Parameters" to "Competing on Practical Implementation"
Over the past three years, the focus has been on large models, chips, and computing power, with the belief that the company with the largest number of parameters or the highest scores in tests would be the most powerful. But the reality is harsh: the models are strong, but companies don't know how to use them, and even if they do, they don't make money.
The new document from the Ministry of Industry and Information Technology aims to address the "last mile" problem in the implementation of AI. The government is no longer just focusing on the giants that create the "engines" (large models) but is now strongly supporting those "application service providers" that can integrate AI into business processes in factories and offices and are responsible for the results.
In simple terms, the government wants to establish a whitelist of "reliable service providers," with the goal of reaching 3,000 by 2027. The company that can turn AI into real profits for businesses will be the next big winner.
---
🔍 In-Depth Analysis: Understanding the Transformation from Five Perspectives
1. Role Reversal: Who Is Doing the "Last Mile" Work?
Previously, we thought AI companies were just those that wrote code or sold servers. But this document redefines the main players as Artificial Intelligence Application Service Providers.
- They are not just simple intermediaries or contractors: They can't just connect an API to create a chatbot and be done.
- They are all-round experts: They need to understand technology (how to tune models), the industry (e.g., automotive manufacturing, financial risk control), and operations (how to maintain systems and ensure data security after deployment).
- The bar has risen: To be included in this resource pool, having just a PPT is not enough. You need real contracts, real customers, real cases (at least 10-30 within three years), and a dedicated team (at least three people with employment proof).
- In plain language: The government doesn't want companies that just talk the talk; it wants those that can deliver results.
2. Solving Four Major Challenges: Why Didn't AI Get Established Before?
The article points out that the difficulty in implementing AI in the past was due to four major issues. This policy aims to fix these:
- Challenge 1: Lack of reliable professionals (information asymmetry). Traditional business owners wanted to use AI but didn't know who truly understood the subject and feared being deceived.
- Solution: Establish a national resource pool, similar to Dianping (a Chinese review platform), to make the capabilities, cases, and qualifications of service providers public, reducing the screening process.
- Challenge 2: One company can't handle everything (fragmented capabilities). An industrial AI project requires computing power, data, software, hardware, and security, which small companies can't manage on their own.
- Solution: Form "application service teams." One company leads, with several upstream and downstream partners working together to provide integrated solutions.
- Challenge 3: Easy to pilot, hard to scale. It's easy to create a demo for Company A, but hard to replicate it for Company B.
- Solution: Develop "small, fast, lightweight, and accurate" product packages. Modularize common functions for quick deployment and reduce the cost of trial and error.
- Challenge 4: Lack of people, money, and data (insufficient resources).
- Solution: Provide computing power credits, open industry data, and encourage universities to train professionals, especially FDE (Frontline Deployment Engineers).
3. Three Major Innovations in Mechanisms: How to Spend Money and How to Work?
This policy introduces three key innovations that directly change the business logic of AI:
- Innovation 1: Tokens Become Like Utilities. In the past, software was purchased once and for all. Now, AI services are paid for based on usage (tokens).
- Note: The policy emphasizes that it's not just about how many tokens are used; it's about the business results each token generates. Spending a lot of tokens without improving efficiency or increasing revenue is a waste.
- Innovation 2: First Purchase, First Use + Risk Compensation. The biggest issue with AI is uncertainty. Companies are hesitant to buy, and service providers are afraid to sell without cases.
- Solution: The government or insurance institutions share part of the risk. If a project fails due to technical reasons, the government or insurance will compensate; if it succeeds, everyone shares the benefits. This provides insurance for innovation, encouraging both companies to take risks.
- Innovation 3: FDE (Frontline Deployment Engineers) as Core Assets. In traditional software, companies define the requirements, and developers create the software. In AI projects, developers don't know the requirements; FDEs are sent to the client site to translate the client's implicit experience into data and code.
- Key: FDEs are not just advanced on-site customer service; their value lies in identifying problems for the first time so that the back-office team can productize solutions for repeated use.
4. Beware of Four Potential Issues in Implementation:
The article clearly identifies potential deviations in enforcement, which investors and practitioners need to be cautious about:
- Don't engage in a "quantity race": To meet the goal of 3,000 providers, local standards may be relaxed, leading to unqualified companies entering the pool. The resource pool must have an exit mechanism to prevent poor quality.
- Don't focus only on the number of cases: Having 30 cases doesn't mean they are all profitable or effective. Consider the renewal rate, payment collection rate, and system stability.
- Don't treat FDEs as mere labor contractors: If FDEs are paid by the number of people without being held accountable for results, it's still the same old approach. They should be paid based on their effectiveness.
- Don't treat token consumption as a performance metric: The government and companies should not use AI just for the sake of using it. Focus on total factor productivity (e.g., manufacturing efficiency, customer acquisition costs in services). If the investment is large but the output remains unchanged, it's an inefficient investment.
5. What Does the Capital Market Think? Who Will Be the Winners?
For investors and listed companies, this document changes the valuation logic:
- Revaluation: Software companies, once seen as low-margin, labor-intensive project-based businesses, may now be valued higher if they have industry expertise, delivery systems, and continuous operational data.
- Winners: It's not the companies with the largest model parameters, but those that have deep expertise in 1-2 vertical industries (e.g., healthcare, manufacturing, finance) and can create a closed loop of "data-model-business."
- Key indicators: Look at recurring revenue (subscription-based income), customer renewal rate, and the number of customers supported by each FDE (efficiency metrics).
- Opportunities for Hong Kong and Overseas Companies: Hong Kong companies can leverage their international financing, compliance, and legal advantages to become a bridge for mainland AI service providers to enter markets like ASEAN and the Middle East.
- **Hong Kong-listed small and medium-sized companies can merge with mainland companies to integrate technology and scenarios, creating platform-based AI+industry companies.
---
💡 Suggestions for Ordinary People and Practitioners
1. If you are a traditional business owner: Don't blindly adopt large models. Start with a small-scale pilot using a provider from the resource pool, focusing on whether they can solve your specific problems (e.g., reducing inventory, improving customer service efficiency), and check for performance-based contract clauses.
2. If you are in the AI industry: A pure technical background is no longer enough. You need to understand the industry and the business. Developing FDE capabilities (problem-solving on-site, translating business needs) will be the core competitiveness in the future.
3. If you are an investor: Be wary of companies that only talk about large models without real delivery cases. Focus on those with deep industry expertise and those transitioning from project-based to subscription/operational revenue models.
In one sentence: In the second half of the AI era, the competition is not about who is the most "intelligent" but who is the most "reliable," who can implement effectively, and who can make sound financial decisions.