Tussle between open and closed AI reshaping global order

The tussle between open-source and closed-source artificial intelligence models in China and the United States is more than a debate over technological paths. It reflects the broader China-US technological rivalry, influencing the underlying architectures, industrial control and global rule-making. The clash between these two approaches could profoundly reshape the landscape of the intelligent age and global technological competition.

In recent years, Silicon Valley giants such as OpenAI, Anthropic and Google have built towering technological advantages through closed-source business models, backed by massive capital investment and vast computing clusters. At the same time, the open-source camp — which includes Chinese technology companies Alibaba, DeepSeek and Zhipu AI and US players such as Meta — has risen rapidly, thanks to more efficient algorithms and open ecosystems.

Broadly speaking, the closed-source approach favored by leading US AI laboratories aligns with Washington's desire to maintain technological dominance and Silicon Valley's capital-intensive model, which hinges on massive computing power and premium pricing. US closed-source models, represented by OpenAI's latest reasoning model GPT-5.6 Sol, Anthropic's Claude Opus 5 and Google's Gemini 3.5 and 3.6 series, use top-end GPUs and sophisticated data engineering to build formidable computing barriers and seek a commanding position at the technological frontier.

By keeping model capabilities behind cloud-based application programming interfaces, US companies have created highly controlled commercial ecosystems. This model can also become a geopolitical instrument. By restricting access to advanced models in the name of "national security", Washington has developed a broader narrative: frontier AI may pose national security risks and therefore should remain concentrated in the hands of a select group of Western companies.

High API subscription fees and dependence on cloud services raise the financial barriers for small and medium-sized companies seeking access to advanced AI. They also leave countries without domestic computing capacity increasingly dependent on foreign digital infrastructure, creating strategic vulnerabilities.

Faced with US restrictions on advanced chips and computing power, Chinese tech companies are increasingly turning to open source as a strategic breakthrough, pursuing an asymmetric path based on extreme algorithmic efficiency and large-scale industrial deployment.

At the algorithmic level, DeepSeek demonstrates that performance comparable to leading Silicon Valley closed models can be achieved at significantly lower cost, challenging the assumption that breakthroughs necessarily require billions of dollars and ever-larger, energy-intensive computing clusters. At the ecosystem level, Alibaba's Qwen 2.5 and related models are among the world's most widely downloaded open-source models, serving as a popular foundation for developers.

A report by the US-China Economic and Security Review Commission highlights two key areas of Chinese advantage: a "digital loop" and a "physical loop". Open-source models lower the cost of AI deployment, allowing them to move rapidly into manufacturing, logistics, robotics and other sectors of the economy. The data and practical demands generated by these applications, in turn, feed back into model development.

This creates a virtuous cycle of efficient algorithms, open ecosystems and industrial validation. Together, these strengths give China's AI sector considerable resilience and the potential to catch up or even overtake competitors despite the constrained access to advanced computing resources.

The China-US contest between open-source and closed-source models is also accelerating a broader rebalancing of global technology, giving the Global South and other economies, including those in Southeast Asia, the Middle East, Europe and Latin America, an opportunity to enhance technological sovereignty rather than remain caught between major powers.

Reuters reported on Aug 14 that the US was preparing to ask dozens of countries to choose between US — and China-led AI cooperation frameworks, with those joining Beijing's initiative potentially excluded from a US-led AI coalition. The move shows how competition over AI has extended beyond technology into the geopolitical arena.

For European and Middle Eastern countries, sending sensitive industrial or government data to cloud servers controlled by Silicon Valley giants poses unacceptable compliance and security risks. Yet developing a large foundation model from scratch requires enormous financial and computing resources.

In this context, high-performance open-source models such as Qwen, DeepSeek and Llama are emerging as digital infrastructure for countries seeking to build their own localized AI capabilities. By fine-tuning these models in local languages and deploying them on private infrastructure, institutions in countries such as France, Singapore and Saudi Arabia can develop more cost-effective and independently controlled AI systems. Open source is thus transforming AI from a strategic technology concentrated in the hands of a few countries into a global public infrastructure.

The divide between open and closed models is also a contest over competing approaches to AI safety, governance and regulation. The closed-source camp has long argued that keeping advanced models inside a "black box" is the best defense against misuse, including applications involving biological weapons or cyberattacks. This argument has also been used to justify technological restrictions and stricter administrative controls.

But as open-source models become increasingly capable, the idea that security can be achieved primarily through secrecy is facing growing scrutiny from academia and the international community. More standards bodies and regulatory institutions are recognizing that excessive restrictions on access can easily turn into barriers that reinforce geopolitical and commercial monopolies. From this perspective, governance built on transparency, auditability and multistakeholder participation offers another path toward trustworthy AI.

Looking ahead, the global AI landscape is unlikely to be defined by complete technological closure or the wholesale replacement of closed-source models with open-source ones. A dual-track competition is more likely.

Closed-source laboratories, particularly in the US, will continue to use their enormous advantages in capital and computing power to pursue high-risk exploration at the uncertain frontier toward artificial general intelligence. Open-source ecosystems, led primarily by China and the wider global developer community, will rely on lower costs, stronger adaptability to real-world applications and millions of developers to become a major engine bringing AI into industries across the economy.

In effect, the competition between open-source and closed-source AI models is breaking down barriers to technological monopoly and accelerating the wider diffusion of AI. As a result, a more multipolar, resilient and competitive global AI order is beginning to take shape.

The author is the deputy director of the Institute of American and European Studies at the China Center for International Economic Exchanges.

The views don't necessarily reflect those of China Daily.

If you have a specific expertise, or would like to share your thought about our stories, then send us your writings at opinion@chinadaily.com.cn, and comment@chinadaily.com.cn.