AI and the Chinese Rice Paddies

While reading Malcolm Gladwell’s Outliers, I came across an effective metaphor for describing the gap—and perhaps the future balance—between American and Chinese AI models.

The answer may lie in the rice paddies.

Historically, Western agriculture followed a highly mechanical logic: to increase productivity and yields, farmers introduced more powerful machinery, cultivated larger areas of land, and invested more capital. More resources produced more abundant harvests.

In China and Japan, the situation was different. Rice farmers had smaller plots and less capital to invest in equipment. To increase their yields, they therefore had to improve the process: understand the soil better, refine their techniques, and work with greater precision and consistency.

Rice paddies offer a perfect parallel with AI today and with the gap between the United States and China.

In the United States, companies such as OpenAI and Anthropic have been able to draw on enormous amounts of capital and much broader access to Nvidia’s most advanced GPUs and sophisticated computing infrastructure. The dominant strategy has been, at least in part, to scale up: more data, more compute, and increasingly larger models.

In China, partly as a result of restrictions on the most powerful hardware, many labs have had to focus more heavily on efficiency: better architectures, more optimized training, distillation, and cheaper inference. Models such as Qwen, DeepSeek, and others are showing that competitive performance can also come from more efficient architectural and technical choices, not only from increasing the resources deployed.

That does not mean scale is irrelevant. Far from it. But when you cannot afford the largest machine, you are forced to become better at cultivation.

The interesting question, then, is not whether Chinese models will catch up with American ones using fewer resources, because on many benchmarks they are already doing so.

The question is: what will happen when the greater efficiency and technical expertise Chinese engineers are developing are combined with comparable access to the “machinery”—that is, to chips and computing infrastructure?

At that point, the advantage will no longer lie only in the amount of computing power available, but also in the techniques and expertise built during the years when resources were scarcer.