SEOUL, March 11 (AJP) - South Korea is moving to build its own AI backbone — an "AI Highway" of massive data centers, specialized semiconductors and autonomous software — as it seeks to avoid falling behind the United States and China in the global artificial intelligence race.
Government officials and industry executives outlined the strategy at a briefing at the Seoul Foreign Correspondents' Club on Wednesday, describing an effort to link large-scale computing infrastructure with next-generation chip development and industrial AI systems.
The initiative, led by the administration of President Lee Jae Myung, combines state investment in computing capacity with private-sector advances in semiconductor design and AI applications.
The government has allocated 10.1 trillion won ($6.8 billion) in the 2026 budget to strengthen the country's AI infrastructure. A core goal is securing 50,000 high-performance computing units — specialized processors needed to run advanced AI models.
Ha Jung-woo, presidential secretary for AI and future strategy, framed the push as a matter of national survival.
"Advanced technology like AI is both economic power and the force that determines national security," Ha said.
While the United States and China dominate global AI infrastructure, Seoul is attempting to build what officials describe as a "Silicon Shield" — a network of large domestic data centers powered by carbon-free energy.
One flagship hub is planned in Haenam, a coastal county selected for its potential to host large-scale solar and nuclear power facilities needed to run energy-intensive AI computing.
The government is also moving to integrate AI more rapidly into military operations. Ha said the defense system — traditionally structured and procedural — is being redesigned for faster adoption of AI technologies.
A new deputy minister-level position has been created within the Ministry of National Defense to oversee AI strategy, while the Defense Acquisition Program Administration is preparing to incorporate AI tools into procurement and operational planning.
"South Korea aims to become one of the world's top four defense powers, and AI will be at the center of that," Ha said.
Shift toward specialized silicon
At the hardware level, the AI boom is pushing the industry toward specialized chips designed specifically for AI workloads.
Graphics processing units, or GPUs, remain the dominant technology for training large AI models, but their heavy electricity demand has become a growing constraint for data centers.
South Korean startup FuriosaAI is targeting this bottleneck with its second-generation neural processing unit, RNGD — pronounced "Renegade" — which has recently entered mass production.
Unlike GPUs, NPUs are designed specifically for AI inference, the stage where trained models process new data and generate responses.
Kang Jee-hoon, chief research officer at FuriosaAI, said the industry is entering what he described as a "power crisis," where computing capacity is increasingly limited by electricity availability.
"The challenge for the industry is to enable more work to be processed with the same power consumption," Kang said.
The RNGD chip uses a proprietary Tensor Contraction Processor architecture that manages on-chip memory more efficiently than conventional chip layouts. According to the company, its PCIe server card operates at around 180 watts while delivering roughly 2.8 times higher throughput than comparable hardware within a standard 15-kilowatt server rack.
"Our goal is to generate more tokens with the same power," Kang said. "Just as computing once shifted from CPUs to GPUs, we want developers to easily adopt our Renegade and next-generation products."
At the software level, the next frontier is "agentic AI" — systems that can independently plan and execute tasks rather than simply respond to user prompts.
LG AI Research is advancing this trend through its EXAONE model. Stanly Jung-kyu Choi, vice president and head of the institute's agentic AI research group, described the system as an "expert AI" designed for specialized industrial applications.
In manufacturing, the system is already being used to optimize naphtha scheduling — the complex logistical planning required for petrochemical feedstocks. In life sciences, the EXAONE Discovery platform has reduced the time required to identify new material compounds from about 22 months to a single day by autonomously analyzing research papers and molecular structures.
Because of the high autonomy involved, LG has established a dedicated AI ethics unit to monitor potential risks associated with the technology.
"We are moving beyond general-purpose models to expert systems that maximize productivity in specialized industries," Choi said.
The institute is also developing K-EXAONE, a national flagship model tailored to the Korean language and local context, which is expected to be deployed across public services by late 2026.
As artificial intelligence evolves into a system of specialized hardware and increasingly autonomous software, South Korea is attempting to build a fully integrated ecosystem.
Officials say the success of that strategy will depend on coordination between government policy, semiconductor innovation and advanced research — an effort aimed at securing the country's place in the rapidly shifting global AI supply chain.
Copyright ⓒ Aju Press All rights reserved.
