SK Group Chairman Chey Tae-won Unveils $1.37 Trillion Plan to Build AI ‘Intelligence Factories’ and Expand Chip Supply
July 1, 2026
SK Group Chairman Chey Tae-won Unveils $1.37 Trillion Plan to Build AI ‘Intelligence Factories’ and Expand Chip Supply
SK Group Chairman Chey Tae-won has laid out one of the most ambitious investment blueprints in South Korea’s industrial history, pledging roughly 2,100 trillion won (about $1.37 trillion) over the next decade to transform the country into a global exporter of artificial intelligence. The announcement, made during a national briefing at the former Blue House on June 29, is significant not only for its staggering scale but for the conceptual framework that ties together two massive spending commitments: approximately 1,000 trillion won ($654 billion) for AI data centers and about 1,100 trillion won ($719 billion) to expand semiconductor production, including a new 400 trillion won ($261 billion) chip cluster in the southwest.
At the heart of Chey’s vision is a redefinition of the data center itself. Rather than a passive storage warehouse, he described the AI data center as an “intelligence factory”—a production plant whose output is intelligence generated by running AI models. This reframing, Chey argued, is the logic that drives both halves of SK’s commitment. “We need to turn Korea from a country that consumes AI into one that exports intelligence,” he said. Under this plan, led by SK Telecom, SK will build a total of 15 gigawatts of AI data centers, starting with 5 GW in regions where power and land are secured—including Ulsan, the southwest, Gangwon, and the central region—constructed in 0.5-to-1-GW blocks for speed. A second phase will add another 10 GW as demand, electricity, water, and memory supply allow. By 2035, Chey said, about $654 billion would be invested through strategic-partner investment, customer occupancy contracts, and project financing, describing the infrastructure as Korea’s potential “AI national backbone” and “the heart that drives robots and physical AI.” The 15 GW target represents an enormous electricity requirement, on the order of many large power plants, which is why the rollout is explicitly gated on grid availability.
To feed these factories, the second half of the plan focuses on memory supply. Chey said SK hynix would sharply increase output, pulling forward completion of its Yongin semiconductor cluster—originally slated for 2045—by 12 years to 2033, with about 600 trillion won ($392 billion) allocated for DRAM production in Yongin and roughly 100 trillion won ($65 billion) for NAND flash in Cheongju. Even with this acceleration, Chey warned that shortages are likely to persist, which is why SK is adding the approximately 400 trillion won southwestern cluster on top of existing plans. “Site selection and infrastructure have to begin now,” he said, noting that the Yongin cluster alone took nine years to establish. The new cluster is part of the 1,100 trillion won semiconductor total. Chey said SK would execute the investments while watching demand, with an average of more than 100 trillion won ($65 billion) a year in domestic investment planned over the next decade, backed by what he called a feasible, risk-aware financing plan.
One of the most striking aspects of Chey’s argument is its restraint. As the head of the company with the largest share of the high-bandwidth memory market, he acknowledged that racing to build is necessary because shortage is dangerous—but he also warned that excessive shortage is dangerous too. “If scarcity drives prices to spike too sharply, it could ultimately shrink both the memory and the AI markets,” he cautioned, making sustainable supply expansion necessary rather than optional. This unusually candid acknowledgment from the supplier with the most pricing power signals that the goal is a market healthy enough to keep growing, not maximal scarcity. All of these are announced plans stretching over roughly a decade to 2035, and Chey was explicit that execution depends on demand, power, water, and financing unfolding as projected. Taken together, they sketch a coherent wager: that the country which builds both the intelligence factories and the memory to run them, and paces the build so the market survives the boom, is the one that gets to export the result.
Source: techtimes
