When the Chinese start-up DeepSeek stunned the world and rattled stock markets in late January with the release of DeepSeek-R1—an open-source generative artificial intelligence model that rivaled the most advanced U.S. models while using far less processing power than was thought necessary—South Korea joined a nervous wave of governments around the world making reflexive moves.
On Feb. 17, the South Korean government banned new downloads of the Chinese chatbot, citing data security concerns. Earlier in February, South Korea’s National Intelligence Service released a statement warning the public against DeepSeek’s data privacy risks and inconsistent answers to questions about the historical legacy of northeast China and the origins of kimchi, issues where online Chinese and Korean nationalists have often clashed in the past.
Spurred by the warnings, South Korean users flooded the app with geopolitically charged questions such as “Whose territory was Goguryeo?”—an ancient Korean kingdom often claimed by China—and “Where did kimchi originate?”
South Korea’s frantic reaction was fueled by geopolitical nerves and a sobering awakening to its own lagging AI competitiveness. Earnest efforts over the past few years to develop homegrown models have staggered under capital constraints and a thin talent pool.
But as the initial frenzy subsides, DeepSeek’s democratization of AI—sharing its blueprint for building a powerful system seemingly without investing billions of dollars in specialized computer chips—is signaling an opportunity for South Korea to enter the frontier AI race.
The highly digitized country appears to be embracing DeepSeek’s breakout success as a wake-up call, as the government, tech giants, and start-ups are now rushing to forge strategic partnerships with the United States and to build homegrown AI models, drawing inspiration from the Chinese start-up’s efficiency breakthrough.
Although South Korea is home to one of the most digitally connected populations in the world, generating vast swaths of data to train AI systems, a vibrant AI ecosystem has yet to bloom there. Tech titans have hesitated to make bold investments, the government took a wrecking ball to science research funding in 2024, and salaries for engineers are uncompetitive. Bridging academia and industry remains a challenge, and computer chips are insufficient to power research. Despite having a high concentration AI talent, a 2024 report published by the Stanford Institute for Human-Centered AI found that South Korea is the world’s third-largest exporter of such talent.
But in the wake of DeepSeek’s debut, South Korea is making moves to expand its entrepreneurial economy. On Feb. 4, Sam Altman, the CEO of OpenAI, visited Seoul to meet with Lee Jae-yong—the chairman of Samsung, South Korea’s semiconductor giant, which is already caught in a trade war between the United States and China. On the same day, Altman also met with Chung Shina, the CEO of Kakao, a South Korean internet conglomerate known for its flagship messaging app, KakaoTalk. The gesture was seen as the start of a possible AI partnership between South Korea and the United States—and the first step of AI diplomacy intended to counter China’s sudden surge.
“On Stargate, there will be many Korean companies that are important contributors to the ecosystem,” Altman said in a joint press conference with Kakao, referring to the $500 billion joint venture between OpenAI, Oracle, and Softbank to build AI data centers, which was announced at the White House and endorsed by U.S. President Donald Trump.
Altman, an ambitious AI diplomat, also stressed that South Korea’s energy, semiconductor, and internet companies made the country an important market for OpenAI, noting that South Korea’s demand for AI was growing “super fast.”
Chung announced that Kakao and OpenAI would jointly develop a new “AI agent”—an AI system that can use computers on behalf of humans—and enhance Kakao’s AI messenger app, Kanana, by integrating OpenAI’s chatbot, ChatGPT. With Kanana trailing far behind leading AI models, Kakao has hatched a catch-up strategy of “model orchestration,” forming strategic partnerships with AI powerhouses and integrating their models into its own products.
While an AI alliance with U.S. companies heralds a promising entry into the global AI race, experts warn that to secure long-term AI competitiveness, South Korea must build a sovereign foundation model—a large language model developed by local engineers and trained on domestic data in order to encode a country’s own cultures and customs. This vision—championed for less than benevolent reasons by Jensen Huang, the CEO of Nvidia, the chipmaker central to the AI boom—is starting to gain traction in South Korea.
Developing a homegrown AI model would not only foster technological and economic autonomy, shielded from the whims or market forces of AI superpowers, but could also bear profound geopolitical consequences. Amid the digital cold war between the United States and China and the global hegemony of Big Tech, as well as growing concerns over censorship and disinformation parroted by AI chatbots, South Korea sees building a domestic vintage of the technology as integral to maintaining its geopolitical standing—and preserving its norms and values.
“Harnessing our own AI technology is imperative, as it is directly tied to state sovereignty,” said Kim Kee-eung, a professor of computer science at KAIST University and the director of Korea AI Research Center. “An AI model developed in another country will inevitably be biased as it is trained on data that reflect that country’s own values,” he said. “If we adopt such a model, the social values of another country could subtly reshape our decision-making agency.”
Many South Korean tech companies share this vision of AI sovereignty. Naver, the internet conglomerate behind South Korea’s biggest search engine—which outpaces Google domestically—has developed a homegrown large language model, HyperCLOVA X, from scratch. This was done using vast troves of data from Naver’s search engine, blogs, and forums, as well as data purchased from encyclopedias, to optimize for the Korean language and reflect supposed Korean culture and values.
But the model has yet to rival leading U.S. models, owing in part to constraints in computing power. Still, when Naver unveiled its AI chatbot, CLOVA X, and an AI search engine, CUE:, in 2023, their performance signaled South Korea’s growing AI prowess.
“It’s difficult for South Korean companies to make comparable investments,” said Ha Jung-woo, the chief of Naver’s cloud AI innovation, pointing to the astronomical infrastructure spending by U.S. tech giants. “But we can make the investments made by DeepSeek,” he added.
Ha is optimistic that South Korea can build a self-reliant AI future, as he believes that DeepSeek’s open-source model has empowered latecomers to the AI boom to make rapid technological advancements even with a modest computing budget.
Start-ups are joining the brewing optimism. “This could open opportunities for start-ups like us,” said Kwon Soon-il, the vice president of the South Korean start-up Upstage, founded in 2020 by a group of engineers from Naver.
In December 2023, when Upstage released Solar, a compact large language model that was fine-tuned from Meta’s open-source model Llama, the model topped the Hugging Face Open LLM Leaderboard—a platform that evaluates and ranks open-source models—signaling that smaller South Korean models are catching up to advanced U.S. counterparts.
“Start-ups in South Korea can use [Deepseek’s] program’s code or analyze its methodology to build their own AI models,” Kwon said.
Still, rather than blindly following the footsteps of DeepSeek, he says that companies should push themselves to pioneer a new methodology that challenges the Chinese start-up’s innovation—to engineer a kind of technological autonomy.
“DeepSeek has been an important catalyst for South Korea,” said Park Tae-Woong, the chair of the civic forum Question for All and the author of the book AI Lecture 2025. “It has shocked us, made us reflect on our weaknesses, and shown us a road map for what to do moving forward. If we devise a national strategy, concentrate our resources, and bridge industry and academia, we can catch up.”
Spurred by a new sense of urgency, the South Korean government is rushing to support domestic AI firms. On Feb. 17, acting President Choi Sang-mok—serving in place of impeached President Yoon Suk-yeol who leads the conservative People Power Party (PPP)—pledged in mid-February to purchase 10,000 GPUs within the year to make South Korea one of the world’s “top three AI powerhouses.” Choi stressed that the frontier AI race was shifting from a battle between companies to one between countries.
The opposition Democratic Party is stepping up as well. “Although we have fallen behind, DeepSeek gives us hope that even late bloomers will have opportunities in the AI industry,” said party leader Lee Jae-myung in a speech given at the National Assembly on Feb. 10. Lee proposed myriad initiatives to accelerate innovation, including building a national AI data center, improving infrastructure for researchers and developers, and establishing an AI boot camp to train industry professionals. “We need a strong government initiative for an AI revolution,” he said.
The legislative legwork is underway to make way for an AI boom. In December 2024, South Korea’s National Assembly passed the AI Basic Act, the country’s first landmark legislation on the technology. The act promotes AI development with various government programs and establishes a regulatory framework that adopts the restrictive approach espoused by the European Union’s AI Act, incorporating its guardrails on transparency, safety, and fundamental rights. While the act is basic in its skeleton, it lays the groundwork for more comprehensive AI governance legislation in the future.
In a world where technological innovation breeds geopolitical prowess, South Korea is racing to spur the next round of AI breakthroughs. But the push for sovereign AI isn’t just an endorsement of techno-nationalism. The move also nods to a kind of techno-pluralism—the idea that a wealth of competing technological systems trained on diverse datasets fosters a world more attuned to its complexities and more resilient to asymmetries and falsehoods.
As AI power centers emerge and shift around the world, so too will their representation of the voices that inhabit it. The better question to ask may not be who leads the race, but how we can build a digital intelligence infrastructure that preserves epistemic diversity.
