"They Blocked China's Chips, and It Became a 'Value-for-Money Monster'..." China's Counterattack, Copying the U.S. and Selling It for Peanuts [China Shock, K-Industry on the Brink | AI] 1

China Projected to Lead in LLMs by 2030

Editor's NoteChinese artificial intelligence (AI) technology is threatening South Korea's industrial ecosystem and national security. Backed by vast capital and infrastructure, Chinese companies are encroaching on sectors ranging from the tech industry to the public and financial sectors, using value for money and open source as their weapons. This is no longer merely a matter of borrowing technology: the risk to software (SW) supply chain security is becoming real, as control over data and national sovereignty could be transferred abroad from the design stage onward. We examine the technological onslaught of Chinese AI and explore survival strategies for South Korea's AI industry, including securing independent AI foundation models to safeguard national data sovereignty.

"U.S. export controls on advanced graphics processing units (GPUs) to China have slowed China's pursuit of the United States, but they have also given rise to a uniquely Chinese AI ecosystem built around computational resource efficiency and open distribution."

As China rapidly takes over global markets with so-called "value-for-money AI models," projections indicate that it will lead the world in large language models (LLMs), a core AI technology, by 2030. While U.S. frontier AI companies operate closed models that are available only through application programming interfaces (APIs), China is aggressively expanding its reach by making model weights available in an open-weight format.

On October 28 (local time), Bloomberg Intelligence (BI) projected in a report that China would lead the world in eight of 17 key industries, including LLMs, by 2030. "China will lead the world in LLMs by 2030," the report said, adding, "The United States and China are likely to share leadership in AI technology, and neither side is likely to gain an overwhelming advantage." (Related article: "China to Lead the World in AI Language Models by 2030, Sharing Tech Supremacy with the U.S.")

China is expanding its developer base with models that emphasize lower costs and openness. Jung Junhwa, a legislative researcher at the National Assembly Research Service, explained, "China has found it difficult to secure large-scale computing resources because of the U.S. government's export controls on advanced GPUs. Concluding that it would be difficult to catch up with early leaders through a closed, subscription-based model, China has chosen a circular approach: attracting global users and data with low prices, then reinvesting them in improving its domestic AI technology." He added, "China's approach is likely to converge on a 'dual structure' rather than a retreat from openness: widely distributing general-purpose models as open-weight models while restricting only the most advanced ones."

U.S. Export Controls Spur More Efficient Computing and Give Rise to a Chinese Ecosystem

As U.S.-China competition for AI supremacy intensifies, companies are also stepping up their efforts to counter one another. On October 12 (local time), Anthropic announced it would actively respond to Chinese AI companies' attempts to replicate its models and take defensive measures. Anthropic criticized Chinese AI companies for attempting "distillation attacks," using large numbers of accounts and automated requests to replicate the capabilities of its AI model Claude. A distillation attack involves making a large number of queries to a competitor's high-performance AI model, collecting its responses, and using them to train a rival model.

Anthropic claimed that Chinese AI research institutes had tried to collect large volumes of Claude's responses to train competing models. It said that customers accounting for about 5% to 7% of its revenue from January through September this year were among those targeted for blocking. Earlier, in February, Anthropic announced that DeepSeek, Moonshot, and MiniMax had generated more than 16 million interactions through approximately 24,000 fake accounts. Anthropic subsequently restricted proxy access and blocked accounts, and has since introduced new technical measures.

As AI emerges as a key strategic asset directly tied to national survival, not just technology, companies are tightening access controls to protect core AI model technology. Tensions are also escalating between openness and technology protection in AI services, as well as between customer acquisition and revenue growth. With China gaining AI model competitiveness in a short time and at low cost, the United States is intensifying its efforts to counter China. The highest-performing AI models are not immediately made available to the public; instead, they are initially distributed on a limited basis to selected institutions vetted by the government.

Hong Gwanhee, chief information security officer (CISO) at LG Uplus, said, "The U.S. government was involved in selecting partners for Project Glasswing for both Anthropic's Mythos Preview and Mythos 5, bringing them under national control." He added, "The distribution of advanced AI models is expanding beyond simple business considerations between companies to encompass political and diplomatic relations between governments and companies, as well as between countries."

Concerns Grow Over Supply Risks for Korean Companies Using Chinese Open-Weight Models

The problem is that Chinese AI is gaining broad ground in the Korean market with nearly free, value-for-money open models. In South Korea, a stable supply of models from either the United States or China is not guaranteed, and domestic companies that have relied on Chinese open-weight models for their low cost could face new supply risks. A recent parliamentary audit sparked controversy after it emerged that the Korea Investment Corporation (KIC), which manages approximately 339 trillion won in assets, was using a language model from China's Alibaba and a search model from a Chinese research institute on the U.S. government's export control list in its internal AI system.

As the use of Chinese AI models grows among both domestic private companies and public institutions, the government is pushing for a transition to homegrown models. Bae Kyunghoon, Deputy Prime Minister and Minister of Science and ICT, said, "The reality is that many companies and public institutions are using Chinese AI models." He added, "Once the results of the third evaluation of our independent AI foundation models are released, we will further expand the use of homegrown AI models."

According to domestic developer communities, the use of Chinese models is widespread. Im Jeonghwan, CEO of Motif Technologies, explained, "Chinese AI companies such as DeepSeek, Moonshot AI, and Alibaba make the weights of their fully trained models available on platforms such as Hugging Face, allowing anyone to run them on their own servers and modify them for specific purposes. This is further expanding their reach, especially among developers."

A developer who requested anonymity said, "Anthropic's Claude Fable 5 is highly satisfying in terms of performance, but it is too expensive." The developer added, "Kimi K3, released by China's Moonshot AI, costs up to 70% less than Anthropic's top models, so quite a few Korean companies are using Chinese AI. However, we are keeping a close eye on recent signs that Chinese AI models may be moving away from their low-price strategy."

This content was produced with the assistance of AI translation services.