
The battle for global dominance in artificial intelligence has split into two contrasting strategies. While the United States is focused on restricting access to high-end computing hardware, China is aggressively proliferating accessible, low-cost AI software—raising new questions about which approach will ultimately lock in global market share.
According to a recent industry report by Hana Securities analyst Baek Seung-hye, U.S. export controls on advanced semiconductors and chipmaking tools have successfully bottlenecked China’s raw computing power. Even Huawei’s newest AI processor released this year reportedly lags behind the computing performance of Nvidia’s H100, a chip introduced in 2022.
Yet hardware benchmarks tell only part of the story. Data from OpenRouter, an aggregation platform that allows developers to route queries across various international AI models, reveals a sharp shift in user behavior. Token usage for Chinese AI models overtook that of American counterparts in April and surged to a 68% share on the platform by November. Although OpenRouter represents a single ecosystem rather than the entire global market, the trend highlights how rapidly Chinese software is gaining traction despite hardware handicaps.
The Open-Weight Trojan Horse Beijing’s advantage lies in affordability and open distribution. By releasing pre-trained models that foreign enterprises can download and host on their own local servers, Chinese tech firms alleviate data privacy concerns while avoiding the massive infrastructure costs of processing global queries themselves.
Analyst Baek warns that this open-source strategy serves a broader long-term objective: ecosystem lock-in. Businesses initially drawn to budget-friendly Chinese models may eventually gravitate toward compatible Chinese enterprise software, humanoid robotics, and, ultimately, Chinese-made chips and cloud infrastructure.
Furthermore, the race carries steep national security implications. Advanced AI systems capable of autonomously detecting software vulnerabilities—such as Anthropic’s "Mythos"—grant a decisive asymmetric advantage to whichever nation develops and deploys them first, prompting Washington to consider expanding its containment policies beyond silicon to encompass AI models and robotics.
The Memory Bottleneck Paradoxically, stricter U.S. sanctions are accelerating China's push for technological self-reliance. Historically, Chinese developers bypassed hardware limits by leasing overseas cloud servers, acquiring restricted chips through gray markets, or using "model distillation" to train local systems on outputs from leading U.S. models.
As Washington moves to close these loopholes, China faces urgent pressure to localize its semiconductor supply chain, particularly Extreme Ultraviolet (EUV) lithography machines and High Bandwidth Memory (HBM). Of the two, memory poses the most immediate hurdle; without sufficient HBM to feed data into processors, even domestically produced AI chips cannot operate at peak efficiency. Consequently, industry observers note that the pace of China’s AI infrastructure expansion in the near term will heavily depend on domestic memory maker ChangXin Memory Technologies (CXMT) and its ability to scale HBM production.
[Copyright (c) Global Economic Times. All Rights Reserved.]




























