Summary of the Key Points
This article discusses the topic of whether China's decision to open up the weights of its AI models represents "decelerationism" or "AI communism," with the central argument being that opening up these model weights does not hinder the advancement of AI but rather reorganizes the profit distribution within the AI industry. It counters the claims by OpenAI executives that such openness will reduce companies' motivation to invest in research and development, pointing out that China's choice is based on practical competitive strategies—such as building a global ecosystem, promoting industry collaboration, and expanding industry applications. The future competition between China and the United States in AI will not solely depend on the capabilities of the models but on a balance between "innovation speed" and "industry penetration rate," as well as who can control real-world scenarios, unique data, and new pricing power.
Detailed Analysis
1. The Concerns about "Decelerationism": What Do Model Companies Fear?
OpenAI executives argue that opening up model weights is a form of "decelerationism," which essentially reflects their concern from the perspective of these companies: "We've invested billions in training advanced models, and if they become publicly available, others can use them freely, eliminating our monopoly profits. How will we have the incentive to continue investing in the next generation?" For example, if a company spends 10 billion dollars on training a top-tier model with the intention of making money from API calls, if the model weights are made public, other companies could download and modify the model for their own use, thereby avoiding paying that company. Model companies might ask, "How long will our lead advantage last? Why would customers choose free or cheaper open models? Will investors still invest in us?" This concern is partially valid, but it only addresses half of the issue—it overlooks the fact that profits within the entire AI industry will be redistributed to other sectors.
2. Openness Does Not Mean "Eliminating Profits," But Rather "Redistributing Them"
Opening up model weights does not mean there will be no profits in the AI industry; it simply shifts them from the "model layer" to other areas such as "application layers, computing power layers, and data layers." For instance, the Android operating system is open-source (similar to open model weights), and Google did not make a fortune from Android itself but instead earned substantial revenue from its app store, advertising, and cloud services. The same principle applies to AI: previously, profits were concentrated in a few model companies (like OpenAI through ChatGPT API fees); in the future, models will become as accessible as basic resources, and profits will flow to companies that can effectively utilize them—those providing computing power, managing industry data, developing intelligent tools, or helping factories/hospitals implement AI. Therefore, the term "AI communism" is an exaggeration; it's not about the absence of profits but rather about the distribution of profits no longer being monopolized by a few companies.
3. Why Does China Choose to Open Up Model Weights? It's Not About Ideology, But About Practical Strategies
Chinese companies are opening up their models not because they want to implement "communism" but as part of a strategic approach to the AI competition with the United States:
- Quickly Building a Global Ecosystem: American companies have global brands (such as OpenAI), cloud platforms (AWS), and operating systems (Android), making it difficult for Chinese companies to compete if they use closed-source models. By making models open, Chinese companies can attract global developers to try out and customize them for lower-cost products, thus quickly accumulating users and reputation.
- Driving the Domestic Industry Chain: China has many domestic chip, server, and cloud platform companies. With open models, these companies can develop adaptations (e.g., running models on domestically produced chips) and grow together.
- Leveraging Application Advantages: China's strengths lie in manufacturing, retail, and logistics. Opening up models reduces the cost of using AI in these sectors, enabling faster integration into factories, stores, and hospitals—these applications generate much greater value than the profits from the models themselves.
- Coping with Technological Embargos: With U.S. restrictions on advanced chip exports, opening up models allows other countries and companies to deploy AI locally, reducing dependence on American cloud platforms and enhancing the autonomy of China's AI ecosystem.
4. The Future of AI Competition: It's About "Industry Penetration Rate," Not Just Model Capability
In the past, competition focused on the size of model parameters and ranking scores; in the future, it will be more complex:
- First Layer: Model capability (which models are smarter)—the U.S. still has advantages in chips and capital, but China is closing the gap through engineering efficiency and an open-source ecosystem.
- Second Layer: Accessibility of models (which models are cheaper, can be used locally, and are compatible with more chips)—open models have a clear advantage here.
- Third Layer: Real-world scenarios and data (who can integrate models into actual business processes to obtain unique data)—for example, medical AI that helps doctors diagnose diseases or retail AI that reduces inventory shortages; the data in these real-world contexts is truly scarce.
- Fourth Layer: National industrial systems (who can integrate models, computing power, and applications)—the U.S. relies on a closed-source model plus chips and cloud services, while China combines open models with manufacturing scenarios and engineering solutions.
The ultimate winner will be the country that balances "innovation speed" and "application scale." The American model can achieve rapid technological breakthroughs, but the Chinese model can make AI more accessible to more companies, thereby improving overall national productivity.
5. What Should Companies Do?
In the era of open models, companies cannot rely solely on having a large model as their competitive advantage:
- Model Companies: They must either develop truly cutting-edge models (a few companies that need to stay ahead), provide low-cost infrastructure (earning money from cloud services and model hosting), create intelligent tools (system solutions for businesses), or delve into specific industries (e.g., using AI in healthcare to improve efficiency or in manufacturing to reduce energy consumption).
- Ordinary Companies: They should focus on scenarios, data, and business processes—use open models to optimize inventory management, collect customer data, and use that data to refine their models in a cycle. The real competitive advantage lies not in the model itself but in the ability to integrate models into business processes and continuously generate data.
Conclusion
China's decision to open up AI model weights is neither a form of "decelerationism" nor "AI communism." Instead, it aims to restructure profit distribution, turning AI from a luxury for a few companies into an essential infrastructure that can be utilized by many. The key to future competition between China and the United States will not be about which country has the best models but about who can integrate AI effectively into the real economy and create greater productivity. What do you think? Is model capability more important, or is industry penetration rate more critical?