Editor's note: Increasing artificial intelligence inference has led to greater demand for computing power than can be supplied. Shen Jianguang, chief economist at major e-commerce platform JD.com, discussed possible solutions to the problem in a commentary in People's Daily. Below are excerpts of the article. The views don't necessarily represent those of China Daily.

A severe shortage of computing power has gripped countries and regions across the world. While the crunch is partly linked to physical bottlenecks such as chip manufacturing and power grid capacities, a more important driver is the massive increase in AI demand, which has outpaced the existing supply capabilities.

This surge in demand stems from a fundamental shift in the way AI applications are being used.

First, AI is evolving from a "helpful assistant" into an "autonomous agent". In the past, AI was mainly being used for simple tasks such as text generation and basic question-and-answer interactions. But today, new AI agents can directly operate computers by writing code, organizing files and collaborating across different software programs.

This evolution from merely "suggesting ideas" to "taking action" has drastically increased the computing power required for a single task, sometimes by several orders of magnitude.

Second, top developers are relying on massive computing power to achieve higher levels of intelligence. To make AI more reliable and accurate, leading global models typically use inference-time scaling technology.

Before generating a response, the AI pauses to analyze, reason and check its work in the background, much like a human would. For example, when tasked with researching an industry and writing a comprehensive report, the AI needs to autonomously browse websites, read financial statements and run analysis tools. This single task can run continuously for hours, consuming millions of tokens.

Third, the commercial application of AI has crossed a critical tipping point. Thanks to breakthrough improvements in stability and accuracy, AI is now delivering tangible returns in sectors such as finance, healthcare and core business operations. As a result, companies are more willing to deploy these models on a larger scale, transforming AI demand from experimental tech trials to an absolute business necessity.

Faced with this challenge, tech companies worldwide are actively seeking solutions. In the short term, optimizing allocation is key, with many AI providers focusing their computing resources on business segments that produce direct revenue.

Chinese companies on the other hand are seeking to improve the output efficiency per unit of computing power through technological innovation as an effective way to alleviate the computing power pressure.

In the long run, expanding capacity and exploring innovations will be equally important. Tech giants worldwide are accelerating the construction of data centers and increasingly purchasing high-end chips. At the same time, more efficient model architectures and next-generation storage and interconnect systems are gradually maturing, which will help relieve the pressure on computing resources.

In the AI era, computing power has become a foundational strategic resource, much like water or electricity. Only by balancing short-term optimization with long-term planning, and aligning tech upgrades with capacity building, can the industry ensure steady growth and maintain a competitive edge on the global stage.