Summary of Key Points
In the early stages of the AI competition, it was thought that the focus was on GPU chips and large-scale modeling technologies (where the United States had a clear advantage, putting us at a disadvantage for a time). However, as the competition progressed, it became evident that the development of AI also relies on underlying support systems such as energy, materials, and industrial infrastructure—areas where we have significant strengths. As a result, the balance of the AI competition is beginning to tilt in our favor.
Detailed Analysis
1. Early AI Competition: The United States Took the Lead with “Chips + Models,” but Why Were We at a Disadvantage?
When AI first became popular, everyone focused on the tangible aspects of technology—GPU chips and large-scale models.
- GPUs as the “Engine” of AI: Training large models requires massive computing power, which is largely dominated by U.S. companies like NVIDIA (for example, the A100 chip used in training ChatGPT). It was difficult for us to obtain sufficient quantities of high-end GPUs, and even if we developed them ourselves, it would have taken time.
- Large-Scale Models as the “Brain” of AI: Companies like OpenAI (GPT series) and Google (PaLM) were among the first to release mature large-scale models, gaining a technological advantage.
Without access to high-quality GPUs, our model training was slow; without leading models, it was challenging to implement AI applications, making us feel at a disadvantage.
2. The Situation Has Changed: AI Requires More Than Just “Technology”; It Also Needs “Logistical Support”
It has become clear that AI is not an isolated system but depends on three critical components for its operation and growth:
- Energy: Training a large model consumes tens of thousands of kilowatt-hours of electricity (equivalent to several decades of a typical household’s usage), and AI servers need a stable, affordable power supply that operates 24/7.
- Materials: Chips require materials like photolithography resists and silicon wafers, as well as rare earth magnets; AI hardware (such as servers and sensors) demands various specialized materials, all of which are fundamental to the AI industry chain.
- Industrial Infrastructure: A complete industrial ecosystem, from material processing to equipment manufacturing and supply chain management, is necessary to assemble these components and ensure uninterrupted supply.
3. Our Advantages in Energy, Materials, and Industrial Infrastructure:
We have several strengths in these areas:
- Energy: We are the world’s largest producer of renewable energy (leading in solar and wind power capacity) and also have stable supplies of traditional fuels like coal and hydropower, resulting in lower electricity costs. This means we can train AI models more efficiently without worrying about energy shortages.
- Materials: We possess over 70% of the global reserves and production of rare earths (essential for AI motors and sensors), as well as sufficient capacity for mid-to-low-end semiconductor materials like silicon wafers and photolithography resists. Although we still need to improve in high-end areas, our infrastructure supports most AI hardware production.
- Industrial Infrastructure: As the only country with a complete range of industrial categories according to UN standards, we can manufacture everything from chip raw materials to AI servers and sensors, boasting a resilient supply chain.
4. The Implications of This Shift: The AI Competition Is a Test of Comprehensive National Strength
This change highlights that AI competition is not just about individual technologies but about the overall strength of a nation.
- We no longer need to focus solely on chip shortages; we can leverage our advantages in energy, materials, and industrial infrastructure to build a competitive edge. For example, using affordable and stable energy to reduce training costs and leveraging a comprehensive industrial base to quickly deploy AI applications.
This shift means we have moved from a position of passive追赶 to an active strategy—while the United States has strengths in chips and models, we have the logistical advantages needed for sustained AI development. Both sides now possess unique strengths.
In short, the AI competition is like a marathon: the United States initially had a sprint advantage, but with our endurance and logistical support, we are gradually catching up. Who will win in the long run depends on overall national strength.