Observations on domestic AI regulation spanning 3 years across 228 posts. The conclusion is simple:The pace of domestic regulation lags behind market demand and is gradually tightening; small and micro teams almost have no choice but to go overseas.
1. Core Judgments
| # | Judgment | Status |
|---|---|---|
| 1 | MJ will definitely be blocked | ✅ Predicted in 2023-04, later verified |
| 2 | Overseas AI services will most likely be blocked | ✅ Continuously verified |
| 3 | Domestic large model filing has a high threshold | ✅ Verified (impossible for small and medium teams) |
| 4 | Domestic to B AI is hard to do | ✅ Verified (guanxi/connections matter more than products) |
| 5 | Domestic to C will inevitably be free | ✅ Verified after 2025-02 |
| 6 | Large companies are more dragged down by regulatory costs than small companies | ✅ Verified |
2. Regulatory Evolution Timeline
graph TD
A[2023-Q1 MJ/GPT 套壳浪潮] --> B[2023-Q2 网信办首批备案<br/>百度/阿里/智谱]
B --> C[2023-Q3 文心一言公测<br/>大模型必须备案]
C --> D[2024-Q1 大模型监管细则<br/>训练数据合规]
D --> E[2024-Q2 网信办新审核机制<br/>提交模型/训练集/语料]
E --> F[2025 大模型企业商业化收紧<br/>免费成为主流]
F --> G[2026-Q1 AI 监管"刷脸"扩大]
3. Risks of Shell-Wrapped GPT / MJ (2023 Practical Lessons)
Wang Kai summarized on 2023-04-25: "Shell-wrapped products should make money as early as possible".
Practical risk points
- Traffic entry points get walled: DNS pollution / domain name blocking
- Paying users lose contact: SMS review + high miss rate → must guide users to add WeChat
- Policy entanglement: Business information checks, Cyberspace Administration coordination
- Qualification issues: Large models not filed + content review not up to standard
Wang Kai's "getting walled contingency plan" at the time
- Draft service agreements well
- Prepare user notification channels (WeChat / QR code automatic login)
- Backup service addresses
- Monitor business registration information
4. Cyberspace Administration Large Model Filing (High Threshold)
First batch of filings(2023-08): Baidu Wenxin / Alibaba Tongyi / Zhipu / Kimi / Kuaishou / ByteDance / SenseTime / SAIC Motor / Meituan (Zeng Dayue approx.) etc.
Thresholds:
- The enterprise must have compliance qualifications
- Submit training data / models / corpora
- Content review mechanism
- Model interpretability
It is completely impossible for small and micro teams to obtain filing → The only options are:
- Shell-wrap large models (overseas versions must connect to OpenAI / Anthropic)
- Connect to APIs of already filed large models (Baidu / Alibaba)
5. The "AI Burden" of Domestic Large Companies
Wang Kai repeatedly emphasized (in posts dated 2023-04-07 etc.):
- Large companies dare not openly connect to overseas APIs
- Content review compliance costs are high
- Policy changes directly affect core business
- Large companies can only use domestic second- and third-tier large model APIs
- Large companies are dragged down by regulatory costs, while small and micro teams instead have a window of opportunity
Large company response strategies
- Build AI products overseas (e.g., Lao Huang's newtype path)
- Build internal AI tools (not released publicly)
- Wait for regulation to become clear
6. Domestic AI Content Ecosystem Compliance
6.1 AI-Generated Content (AIGC) Compliance Milestones
- AI generation must be labeled
- Cyberspace Administration policy against AI abuse for SEO
- AI writing products must clearly indicate AI generation
6.2 AI Detection Offense and Defense (Wang Kai's focus 2024-2025)
- Google cracks down on AI-written content for SEO
- AI papers / articles are detected by schools / journals
- AI-assisted cheating in job applications, enterprises use AI for anti-cheating
- Authenticity confrontation in AI painting
- AI fraud offense and defense
Wang Kai's judgment: A giant will definitely emerge in the AI security / detection field.
6.3 Large Model Watermarking (Planned)
- Large models will be required to generate unique identifiers
- Countering AI content abuse
- But open-source models can still bypass it
7. Why Domestic to B AI Is Hard to Do
Wang Kai's summary on 2024-06-19(Medical AI case):
- Domestic hospitals prioritize guanxi/connections and background
- Projects are not just about AI capability, but more about channels
- Legal / medical categories require qualifications, small and micro teams cannot enter
- Industry knowledge is not in public data (doctors' medication lists / clinical pathways)
Suggestions for domestic to B AI entrepreneurship:
- Do not enter heavily regulated industries such as healthcare / law
- Choose "AI + sales" categories that do not require qualifications (Zhu Xiaohu style)
- Sell AI customer service / AI sales leads to small and medium enterprises
8. Why Domestic to C AI Will Inevitably Be Free
Wang Kai's epiphany on 2025-02-13:
- Token costs drop 90% per year → Per-user cost approaches zero
- Domestic free model competition → Any charging will be immediately disrupted
- DeepSeek milestone: Open source + free + high quality → entire industry reshuffle
- Big tech companies cover the costs → Single-user ARPU is negligible in their total revenue
Result:
- Mainstream domestic AI products will never be able to charge again
- Overseas players like OpenAI / Claude will also follow suit with free offerings
- "Pay-for-performance" will never work in the Chinese market
9. The "Benefits" of Domestic Regulation for Small and Micro Teams
4 marginal dividends observed by Wang Kai:
- Niche segments that large companies dare not enter
- Gray areas of cross-border data interaction
- AI security / detection peripherals
- Enterprises are reluctant to use big company AI (due to data security concerns) → small tools have opportunities
But All marginal dividends are short-term, in the long run small and micro teams still must go overseas.
10. Policy Warning Milestones (Repeatedly Followed by Wang Kai)
| Milestone | Content | Impact |
|---|---|---|
| Cyberspace Administration large model filing | Quarterly updated list | Determines which large models are usable |
| Ministry of Industry and Information Technology AI application specifications | Irregular | Determines product compliance |
| Cybersecurity review | Regulatory tightening | Data outbound transfer |
| Personal Information Protection Law | Ongoing | C-end AI product compliance |
11. Domestic vs Overseas AI Entrepreneurship Comparison
| Dimension | Domestic | Overseas |
|---|---|---|
| Large model selection | Must be filed ones (Baidu/Alibaba/Zhipu etc.) | Any of OpenAI / Anthropic / Google |
| Data collection | Strict compliance + user authorization | Relatively loose (GDPR is stricter but controllable) |
| Content review | Mandatory + real-time | Relatively loose |
| Payment | Real-name + risk control | Stripe etc. are friendly |
| Small and micro team startup | High compliance costs | Low compliance costs |
| Large companies | Heavy burden | Burden is controllable |
12. Wang Kai's "Domestic Strategy"
Wang Kai's actual operations in the last 2 years:
- Do not make to C products domestically(except for the transition of Squirrel Quick Look)
- AI smoking cessation / AI investment tools: Focus on overseas, build brand presence domestically
- KeepRule: Focus on overseas
- Large model selection: Fully use overseas ones (Claude / GPT / Gemini)
- Payment: Mainly Stripe / overseas cards
13. Observations on Special Domestic Regulatory Milestones
New dynamics emerging in 2024-2026
- Douyin / Qianwen AI entry attempts(System-level entry grab)
- AI education tightening(K12 training restricted)
- AI medical access not liberalized
- AI investment / financial regulation is strict(Brokerages cannot provide AI-generated advice)
14. Wang Kai Says
"99% of small and micro teams doing domestic AI entrepreneurship should choose to go overseas."
"Domestic to C will inevitably be free, the business model is completely different."
"Domestic to B is a market where guanxi/connections matter more than products."
"Large companies are more dragged down by regulatory costs than small teams."
"AI security / detection peripherals are a huge opportunity in the long run."
References
- Wang Kai's Entrepreneurship Methodology and Business Judgment Framework in the AI Era - Entrepreneurship Methodology (Why Not Choose Domestic)
- Wang Kai's Overseas Promotion SOP and Overseas Multi-Account Matrix Practice - Overseas Promotion (Regulation-Driven Overseas Expansion)
- Wang Kai's AI Industry Judgments and Large Company Competition Landscape Observations - Industry Judgments (Large Company Regulatory Costs)
- Wang Kai's 1000+ AI Product Idea Library and Business Model Analysis - Product Ideas (Opportunities Within Compliance Boundaries)
- newtype AI Business, Industry and Investment Observations - Lao Huang's newtype Overseas Solo Entrepreneurship Comparison