The debate over whether China can catch up to the United States in artificial intelligence (AI) misses the real story. A leading Chinese AI researcher recently stated there's less than a 20% chance any Chinese company will surpass a top US AI firm in the next three to five years, but that statistic obscures a more fundamental question: does innovation thrive under scarcity or abundance? The answer reveals two structurally different approaches to building AI systems, each with distinct advantages and blind spots .

Why Computing Power Tells Only Half the Story?

The numbers are stark. The United States controlled an estimated 74% of global AI computing power in mid-2025, compared with China's 14% . That gap represents what researchers describe as "one to two orders of magnitude" difference in raw computational resources. For context, this means American AI labs have roughly 10 to 100 times more computing capacity than their Chinese counterparts.

This disparity has real consequences. US laboratories can allocate substantial computing capacity to both next-generation research and product deployment simultaneously. Chinese labs, by contrast, are stretched thin. Most of their computing resources go toward delivering products that customers demand right now, leaving little room for exploratory research into future breakthroughs .

"Does innovation happen in the hands of the rich or the poor?" asked Lin Junyang, a technical leader who previously worked on Qwen, one of China's most capable open-source AI models under Alibaba.

Lin Junyang, Technical Leader, Qwen/Alibaba

Lin's question reframes the entire competition. It's not simply about who has more computing power; it's about how different resource constraints shape the way researchers approach problems and develop solutions.

How Two Different Innovation Models Are Emerging?

Compute scarcity and compute abundance have created two distinct innovation ecosystems, each with unique strengths and weaknesses. Understanding this divergence is essential to interpreting what the 20% probability actually means for the future of AI development .

• Abundance-Driven Innovation (US Model): With 74% of global computing power, American labs can afford to run multiple experimental approaches simultaneously, allocate resources to speculative research, and maintain parallel product lines. This approach encourages exploration and risk-taking because failure in one direction doesn't threaten the entire operation.
• Scarcity-Driven Innovation (China Model): With limited computing resources, Chinese researchers must optimize every calculation and make strategic choices about where to invest their capacity. This constraint forces efficiency and creative problem-solving, but leaves little room for the kind of exploratory research that sometimes yields breakthrough discoveries.
• Governance Implications: These two models have different implications for how governments might regulate AI development, how companies allocate resources, and what kinds of innovations emerge from each ecosystem.

The US approach resembles a venture capital model: invest broadly, expect some failures, and scale the winners. The Chinese approach resembles a lean startup model: optimize ruthlessly, focus on what works, and avoid waste. Neither is inherently superior; each produces different types of innovation with different blind spots .

What Does the 20% Probability Really Mean?

When Lin Junyang stated there's less than a 20% chance of a Chinese company surpassing a leading US AI firm in the next three to five years, commentators focused on the number itself. Some treated it as evidence that America's lead is insurmountable; others dismissed it as overly pessimistic. Both reactions missed the deeper insight .

The 20% figure reflects the current resource gap and the structural advantages of abundance-driven innovation in the near term. But it doesn't account for how scarcity-driven innovation might produce unexpected breakthroughs, how geopolitical factors might shift resource allocation, or how different definitions of "surpassing" might apply. A Chinese company might not build the largest language model, but it could create a more efficient one, a more specialized one, or one better suited to specific markets .

The real competition isn't a single race with one finish line. It's a divergence into two different technological paths, each optimized for different constraints and producing different kinds of innovation. Understanding this distinction is crucial for policymakers, investors, and technologists trying to anticipate where AI development is heading.

The question isn't whether China will win the AI race in the next five years. The question is what kind of AI innovations emerge when one ecosystem has abundant resources and another must innovate under constraint. History suggests both approaches produce valuable breakthroughs, just in different domains and at different times.