Summary of Key Points
Chinese AI companies are accelerating their international expansion, but tax compliance has become a major obstacle. Due to the lack of unified industry regulations, unique business models, and significant differences in tax environments across countries, risks are hidden in the details of daily operations, which can easily emerge during expansion or initial public offerings (IPOs). This article breaks down the tax challenges faced by AI companies going global from four key aspects: international structure design, taxation of data assets, risks associated with permanent establishments, and classification of cross-border income, and provides practical recommendations.
Detailed Analysis
#### 1. International Structure Design: Don’t Just Focus on “Low-Tax Areas”; Real Business Is the Foundation
There are two main structures for AI companies to adopt when entering international markets. However, with increasingly strict global tax regulations, relying on “shell companies” for tax avoidance is no longer effective:
- Red Chip Structure: Suitable for companies planning to list overseas—set up a listing entity in the Cayman Islands, use Hong Kong as an intermediate holding layer, and have domestic companies handle research and development (R&D) and local operations, while overseas companies manage user services.
- Direct Investment Structure: Suitable for companies that raise funds domestically—the domestic parent company directly establishes sales and service companies abroad, with core technology, computing power, and intellectual property remaining in China.
Key Constraints:
- Substantive Presence Required: Regions like the Cayman Islands now require companies to have physical offices, full-time employees, and independent decision-making capabilities; otherwise, they may not only miss out on tax benefits but also face anti-tax avoidance investigations in China.
- Tax Agreements: Some low-tax areas do not have tax agreements with many countries, meaning high withholding taxes (e.g., on dividends and royalties) must be paid when funds are repatriated.
- Global Minimum Tax: Large AI companies with annual revenues over 750 million euros are subject to a minimum tax rate of at least 15% regardless of their location. The traditional practice of shifting profits to low-tax areas no longer works; tax structures must align with actual business activities.
It is also important to clarify the division of responsibilities: domestic companies should focus on R&D, while overseas companies handle sales and services. Profit distribution should be reasonable (e.g., domestic R&D efforts should be compensated with a small profit margin) to avoid being accused of profit transfer by tax authorities.
#### 2. Taxation of Data Assets
Data is crucial for AI, but there are no uniform standards for how it should be taxed, which can lead to misunderstandings:
- Training Data: If data is purchased or licensed, the price must be reasonable, and contracts, invoices, and delivery records must match. For data shared between related companies, the nature of the transaction (whether it’s considered a “service fee” or a “data fee”) depends on who performs the work, bears the risks, and creates the value.
- User Data: Data generated by users during service usage is embedded in the product (e.g., conversation data from AI chatbots) and cannot be sold separately. If this data is transferred overseas for use by foreign companies, the profit distribution must be properly structured to avoid being deemed a profit transfer.
Risk Points: The issue is not just the price but the nature of the transaction (e.g., whether it’s described as a service or a license). It is recommended to provide background information and pricing rationale to ensure that the flow of data aligns with contracts and invoices.
#### 3. Risks Associated with Permanent Establishments
A “permanent establishment” refers to a fixed operational presence or agent abroad, which may require payment of corporate income tax. Although AI companies use cloud services and do not have physical offices, they can still face risks:
- Cloud Services: Providing standardized API calls (e.g., general AI models) without control over local servers usually does not qualify as a permanent establishment. However, renting dedicated servers, managing local data centers, or using local cloud resources for exclusive services may result in such classification.
- Agent-Type Permanent Establishments: Overseas agents or teams that sign contracts on your behalf or facilitate transactions can also be considered permanent establishments, leading to tax obligations.
Mitigation Strategies: Assess control over servers, personnel deployment, and service delivery methods in advance. Ensure all contractual and operational aspects are well-defined to avoid hidden risks.
#### 4. Classification of Cross-Border Income
The tax implications vary significantly depending on the type of income. Tax authorities focus on the “substance” of the transaction, not just the name:
- Royalties: Have the highest tax rates (e.g., 20% withholding in some countries). If customers can access and modify your models or relicense them, it may be classified as royalties.
- Technology Service Fees: Lower tax rates (some countries exempt these fees). If customers only use your cloud services without accessing core technology, they may be categorized as such.
- Commercial Profits: If there is no permanent establishment locally, no withholding tax is required. For example, if an overseas company merely sells services without a fixed presence, profits belong to the domestic company.
Precautions: Consult with tax lawyers before signing contracts to clarify customer rights (e.g., whether they can modify or relicense models) and the nature of your services. Standardize contract terms and billing to accurately classify income from the outset.
Conclusion
AI companies going global represent more than just a technological export; it also tests their ability to comply with global governance standards. Integrating tax compliance into business decisions (e.g., structure design and contract wording) is essential for sustainable and long-term success. After all, compliance is not a cost; it is a critical safeguard for long-term development.