The Real Contradictions in AI Development
Many people believe that AI technology is important, and they are right—it is indeed very important. However, the core contradiction in the current AI industry has never been a technical issue, but rather overinvestment. Any industry that falls into a vicious cycle of overinvestment will generate systemic risks, much like the real estate bubble of the past—while it’s certainly unacceptable for people to be homeless, when the entire society engages in real estate speculation, it inevitably leads to massive misallocation of resources.
This principle is identical to that of the dot-com bubble era. When the dot-com bubble burst in the early 2000s, residential broadband technology had only just begun its commercial rollout in 1998 and 1999. It wasn’t until the widespread adoption of mobile internet that internet technology truly matured on a commercial level. Only then did various cloud service technologies gradually develop, and internet companies truly begin to generate profits on a large scale.
Return on Investment Characteristics of Underlying Technologies
The more important an underlying technology is, the less it should be subject to speculation. From basic economic principles, the more fundamental the technology and the more basic the need, the greater the required supply. The electricity revolution, the computer revolution, and the internet revolution—all of these technologies were revolutionary in their time, and their importance to society as a whole is self-evident. Without the widespread adoption of electrical infrastructure, computer hardware, and communication networks, AI would never have had the conditions to emerge.
However, once these foundational technologies truly mature, the market inevitably enters a state of perfect competition, and profit margins decline rapidly. Therefore, the greater the supply of foundational technologies, the thinner the profit margins become in the long run. The current appearance of intense demand for AI in the U.S., with extremely high profit margins, is essentially due to insufficient market competition; the core reason for the supply shortage is monopoly.
The Essence of the U.S. Monopoly Logic
The U.S. deliberately restricts China’s purchase of lithography machines, limits China’s chip production, and even restricts other countries from using Chinese chips and AI services—all with the fundamental aim of maintaining high profit margins for U.S. companies through its monopoly position.
This is easy to understand using railroads as an example: if a capitalist controls the nation’s rail network, as long as the railroads remain functional—no matter how severely the equipment has deteriorated—the capitalist will absolutely refuse to spend money on upgrades. The infrastructure in many Western countries is severely outdated for this very reason—as long as profits can still be made, there is no incentive to update equipment. The ideal scenario is to maintain a low-intensity supply gap, where demand slightly exceeds supply. This prevents other capital from entering the market while preserving strong pricing power. A low-intensity supply gap combined with oligopoly is the capitalists’ favorite “effortless profit” model.
Analysis of Capitalist Behavior
The sudden high priority the U.S. has recently placed on nuclear power plant construction stems fundamentally from the massive power shortage caused by the rapid development of AI. Without the surge in electricity demand driven by AI, capitalists would have no interest in investing in nuclear power plants. Although nuclear power plants provide a stable power supply, their construction requires extensive safety measures, resulting in low returns on investment and an excessively long payback period—under normal circumstances, capital would not even consider them. Only when the power shortage becomes so severe that AI cannot function without electricity do capitalists perceive a profit opportunity and become willing to invest in nuclear power plants.
This short-sighted nature of capital often results in emerging technologies bearing an excessive burden of hidden costs. The ultimate consequence is that the cost of scaling AI from 1 to 100 in the United States will be significantly higher than in China. For example, if Microsoft wants to popularize AI, it must invest in nuclear power plants to solve the power problem. These exorbitant investment costs will ultimately be passed on to consumers, including the currently outrageously priced NVIDIA chips.
Cost Differences in AI Development Between China and the U.S.
China’s infrastructure development is driven by a nationwide system that plans ahead. Essential underlying resources for AI development—such as electricity, networks, and computing power—have long been widely available, with ample supply and low costs. In contrast, U.S. infrastructure is capital-driven; investments are made only where profits can be made, and areas without profit potential are never developed in advance. This difference means that during the AI adoption phase, China’s implementation costs will be far lower than those in the U.S.
Many people focus only on the U.S.’s current lead in large AI model technology, yet overlook the fact that technology must ultimately be put into practical use. The application phase is a contest of cost and efficiency. China possesses the world’s most comprehensive industrial chain and the lowest infrastructure costs; these advantages will gradually become apparent during the large-scale adoption phase of AI.