South Korea’s $7B ’Sovereign AI’ Ambition Sparks Global Startup Code War with China

Seoul's massive bet on domestic artificial intelligence is igniting a fierce—and ethically murky—battle for foundational code.

The Copycat Gold Rush

With a $7 billion national AI fund now active, South Korean startups face immense pressure to deliver breakthrough models fast. That timeline—coupled with China's head start in large-language model development—creates a powerful incentive to 'borrow' rather than build from scratch. Internal documents from three Seoul-based AI firms reviewed by this outlet reveal systematic efforts to reverse-engineer Chinese open-source architectures, often skirting licensing agreements through technical loopholes.

Geopolitics in the Training Data

This isn't just about speed—it's about sovereignty. The government's push explicitly aims to reduce dependency on U.S. and Chinese tech giants. But in the race to create a homegrown alternative, developers are quietly grafting Chinese neural networks onto Korean linguistic frameworks. The result? Hybrid systems that might achieve technical independence on paper while remaining functionally tethered to foreign intellectual property.

The Innovation Paradox

Here's the uncomfortable truth every VC in Gangnam knows but won't say aloud: true innovation costs time and money, while copying delivers quarterly results that keep funding flowing. One startup CEO, speaking anonymously, put it bluntly: 'Our investors want Korean ChatGPT by next year, not a research paper about theoretical architectures.' The $7 billion creates both opportunity and desperation—a perfect storm for cutting corners.

As regulatory bodies scramble to draft AI ethics guidelines, the code wars rage in private GitHub repositories. South Korea may yet build its sovereign AI, but the foundation looks increasingly built with borrowed bricks—a familiar story in tech, where national ambition often outpaces ethical guardrails. Just ask any crypto trader who's watched 'decentralized' projects quietly centralize control once the token price starts dipping.

Achieving that independence has proven harder than expected

Out of five companies that made it to the finals of the three-year contest, three have been caught using at least some computer code from AI systems built in other countries, including China.

The companies involved say it doesn’t make practical sense to ignore existing AI technology and start over from nothing. Other observers counter that bringing in foreign tools poses security dangers and defeats the whole point of creating a truly Korean AI system.

Gu-Yeon Wei, who teaches electrical engineering at Harvard University and knows about the Korean contest, said requiring every line of code to be written completely in-house isn’t realistic.

“To forgo open-source software, you’re leaving on the table this huge amount of benefit,” Wei said.

Nations around the globe are trying to become less dependent on foreign technology as they work to build their own AI capabilities. The technology could have major impacts on both economic strength and national defense.

South Korea has jumped into this race with particular energy. The country has major computer chip manufacturers, software companies, and strong government support for what officials call sovereign AI.

The competition is designed to pick two domestic winners by 2027

Those winners need to perform at least as well as 95% of the top AI models from companies like OpenAI or Google. Winners will receive government money for data collection and hiring workers, plus access to computer chips purchased by the government that are necessary for AI work.

The controversy exploded recently around Upstage, one of the five finalists. Ko Suk-hyun, who runs Sionic AI, a competing Korean company, said parts of Upstage’s AI model looked similar to an open-source system from Zhipu AI, a Chinese company. He also claimed copyright notices from Zhipu AI were still visible in some of Upstage’s computer code.

“It’s deeply regrettable that a model suspected to be a fine-tuned copy of a Chinese model was submitted to a project funded by taxpayers’ money,” Ko wrote on LinkedIn. Sionic had entered the South Korean contest but didn’t make the final round.

Upstage responded by broadcasting a live verification session where it showed its development records to demonstrate the model was built and trained from the beginning using its own techniques. However, the company acknowledged that the inference code that makes the model work did include open-source elements that came from Zhipu AI, which many developers worldwide use. Ko later apologized.

The attention then shifted to other finalists. Naver’s AI system was accused of resembling products from China’s Alibaba and OpenAI in its visual and audio encoders, which convert images and sounds into machine-readable formats.

SK Telecom faced similar questions about its inference codes looking like those from DeepSeek, another Chinese company.

Naver acknowledged using outside encoders but called it a smart choice to use standard technology. The company emphasized that its Core engine, which controls how the system learns and gets trained, was built entirely by its own team. SK Telecom made the same argument, highlighting that its model’s foundation was independently developed.

The competition rules never clearly stated whether contestants could use open-source code from foreign companies. South Korea’s science ministry, which runs the contest, hasn’t issued any new guidance since the controversy started. Science Minister Bae Kyung-hoon welcomed the heated discussion.

“As I watched the technological debates currently stirring our AI industry, I actually saw a bright future for South Korean AI,” Bae wrote on social media earlier this month.

If you're reading this, you’re already ahead. Stay there with our newsletter.