appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Meredith Chen

The global AI race is no longer just about chips or models. Increasingly, it is also about data – namely, the language resources that feed and train them.
Beijing has identified that layer as a new strategic frontier, accelerating the buildout of a national ecosystem of Chinese-language data for artificial intelligence (AI). The push covers everything from foundational text corpora and high-quality training datasets to technical standards and governance frameworks.
Language data has emerged as a “more winnable” arena for China in the global AI race despite intensifying US restrictions on advanced chips and technologies, analysts say, with continuing challenges over data scarcity, quality, standards and governance.
The drive comes amid China’s growing conviction that language resources are no longer simply research assets but strategic infrastructure that underpins the country’s AI ambitions, digital governance and cultural influence.
The efforts have also become increasingly visible in recent months.
On July 5, the Communist Party newspaper Guangming Daily devoted an entire page to the topic, with three articles stressing its strategic significance and calling for faster and more coordinated progress.


The urgency stems from a stark imbalance in the world’s language resources available for AI training.
Large language models learn by ingesting vast amounts of data, and the availability, quality and diversity of language resources have become increasingly important in shaping AI performance.
English overwhelmingly dominates the data used to train most popular models in the world, while Chinese remains structurally under-represented.
According to a 2024 report by AliResearch, Alibaba’s research institute, English accounts for nearly 60 per cent of global web content, while Chinese makes up just about 1.3 per cent, even though it is spoken by more than a billion people.
Alibaba owns the South China Morning Post.
That imbalance gives today’s leading AI systems a “native tongue”, and those who do not speak it are being left behind in subtle and consequential ways, according to Wesley Wu-Yi Koo, an assistant professor of management and organisation at Johns Hopkins University.
Koo said past research had shown that some Chinese open-source models relied heavily on English-dominated web datasets, inheriting English-centric nuances and biases that “might not be suitable for Chinese usage contexts”.
His own research found that AI-generated content in Chinese and Arabic consistently scored lower than English across all quality dimensions, leading non-English speakers to produce “less actionable and less creative” output, with the gap most pronounced in technical tasks.
“A model’s native tongue is a deliberate, consequential allocation decision, and countries that don’t make that investment will accept structural quality deficits they may not fully see,” Koo said.




Chinese AI developers have already begun responding.
Chinese-language data made up more than 60 per cent of the training data used by most Chinese AI models, with some models reporting shares of up to 80 per cent, according to the National Data Administration in August last year.
Official guidelines announced in 2025 on strengthening the construction of “digital Chinese” called for promoting the digitalisation of Chinese-language resources and expanding Chinese-language data resources, with the aim of establishing a national corpus and strategic language resources information database by 2027.
By 2035, China hopes to have significantly expanded the presence of the Chinese language in global digital and generative AI scenarios.
Koo said Beijing now viewed data as “the domain where it can compensate for its chip disadvantage”.
As US export controls keep constraining China’s access to advanced semiconductors, language data had become “a more winnable territory”, supported by China’s vast domestic user base and the state-coordinated capacity to mobilise data across enterprises and institutions that few other countries could match, he added.
“I’d argue data is becoming more decisive than chips in some cases.”
As model architectures converge and open-weight models make advanced AI designs more widely accessible, high-quality language resources may become a more durable source of competitive advantage because the depth of a language’s technical literature and domain-specific corpora cannot be replicated easily, according to Koo.
Building high-quality language resources was far more than data collection, he added, saying it required decisions over access, processing, standards and governance.
In May, China’s official People’s Daily Online launched a media corpus service platform designed to help media organisations transform their content into AI-ready datasets through data processing, standardisation, and copyright and security management.
High-quality news data is an important component of language resources, as professionally produced journalism provides well-sourced and informative materials for AI systems.
According to analysts, while the English ecosystem benefits from decades of organically accumulated digital footprint – from scholarly literature and Wikipedia to programming forum posts – many high-value Chinese knowledge resources are still difficult to access or have yet to be systematically incorporated into AI-ready datasets.


Governance of language resources remains a major challenge, Tsinghua University professor Sun Maosong and assistant researcher Kong Cunliang wrote in Guangming Daily.
They said China still needed a more coherent governance framework to address issues including copyright and data sharing, fragmented technical standards, and privacy and security risks.
Technical standards – including character encoding, semantic annotation and multimodal data integration – had become increasingly important in global AI competition, wrote Rao Gaoqi, an associate professor at Beijing Language and Culture University, also in Guangming Daily.
Falling behind in foundational standards for Chinese-language resources would have direct implications for “national cultural security, data sovereignty and China’s ability to shape international standards”, Rao said.
“If these core standards are controlled by others, China risks losing ground in global digital governance and limiting the reach of Chinese culture in the digital space.”
Koo said governance mode was one of the main differences between Chinese- and English-language AI ecosystems, which faced “different trade-offs”.
Chinese-language resource development involved “ideological curation” and not just technical curation, he added, as Beijing required training data to adhere to “socialist core values”.
Meanwhile, governance in the English-language AI ecosystem was shaped more by constraints such as copyright litigation, transparency obligations and consent concerns.
“These operate through litigation and markets rather than state-directed content filtering, leading to a more pluralistic data landscape,” Koo said.
As AI increasingly becomes the interface through which people access knowledge, solve technical problems and make decisions, differences in language resources risk translating into differences in economic opportunity and technological capability, he warned.
Non-English speakers also bore a higher safety risk, Koo said, with research showing models produced more harmful content and followed instructions less faithfully in low-resource languages, yet major international AI safety frameworks scarcely addressed multilingual safety.
As many governments around the world launch sovereign AI initiatives to strengthen their domestic capabilities, a national model alone would not be enough, he added.
“AI is often touted as a technology that levels the playing field,” Koo said. “My research suggests that, cross-lingually, it may do the opposite unless there is deliberate, sustained investment in language-inclusive models and data.”


Trade truce extension aside, was AI the real Xi-Trump summit breakthrough?

How Chinese firms can avoid a Europe-style backlash in Southeast Asia

How Alibaba is using AI to revamp its mapping app – and take on Meituan

China’s AI-led transformation is taking shape. Why are some economists urging caution?

Artificial intelligence (AI)

Why data may become China’s most durable advantage in the AI race

China sees language resources as strategic infrastructure that underpins the country’s AI ambitions. Photo: Shutterstock

Will AI kill English classes? Chinese university to drop them as tech booms

Shenzhen University of Advanced Technology in southern China has announced it will do away with standard English courses in favour of classes that emphasise cross-cultural skills. Photo: SUAT

Real-time translation has become “impressively effective”, according to SUAT provost Zhao Wei. Photo: WeChat

US says Nvidia’s H200 exports to China remain ‘trivial’ despite approvals

A top US official told Congress on Tuesday that "very few" Nvidia H200 chips have been shipped to China or Hong Kong to date. Photo: Reuters

AI homework tools cut exam scores by 20%, study of 26,000 Chinese students finds

By comparing AI users with non-users, researchers from Sweden and Hong Kong found a clear gap: AI boosted short-term efficiency but harmed long-term learning. Photo: Shutterstock

Global AI boom sees China’s chip exports nearly double in first half of year

A staff member tests chips in a workshop at the Zhangjiang Hi-Tech 895 Incubator in Shangahi’s Pudong New Area on July 8. Photo: Xinhua

A staff member tests chips in a workshop at the Zhangjiang Hi-Tech 895 Incubator in Shanghai’s Pudong New Area on July 8. Photo: Xinhua

China works on AI safety benchmark as regulators target large model risks

Chinese authorities have started building a safety benchmark to evaluate artificial intelligence models. Photo: Shutterstock

News

China

Politics

Accelerating buildout of national ecosystem of Chinese-language data has emerged as a ‘more winnable’ arena amid US tech restrictions


Accelerating buildout of national ecosystem of Chinese-language data has emerged as a ‘more winnable’ arena amid US tech restrictions.


United States

Johns Hopkins University

Tsinghua University

Alibaba

Why data may become China’s most durable advantage in the AI race

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Accelerating buildout of national ecosystem of Chinese-language data has emerged as a ‘more winnable’ arena amid US tech restrictions

Accelerating buildout of national ecosystem of Chinese-language data has emerged as a ‘more winnable’ arena amid US tech restrictions