Korea Commits $919B to 18.4GW AI Datacenter Buildout by 2035
TL;DR
- Korea's plan targets 8.4GW of new datacenter capacity by 2029 and 18.4GW by 2035, backed by a $919B commitment announced in July.
- Phase one splits build-out across SK Group (5GW), GS Group (2.4GW), and Naver (1GW), with Samsung planning an Nvidia site of more than 50,000 GPUs.
- Sub-30-person Motif Technologies, with $17M raised and just 768 B200s, ranks 10 points ahead of Upstage on the Korean tournament leaderboard.
South Korea has committed to a $919B datacenter buildout targeting 8.4 gigawatts of capacity by 2029 and 18.4 gigawatts by 2035, per a SemiAnalysis breakdown of the country's sovereign AI program. The first phase splits 5GW to SK Group, 2.4GW to GS Group, and 1GW to Naver.
Alongside the physical buildout, Seoul is running a foundation-model tournament to pick a national champion. SemiAnalysis lists the surviving cohort as "Upstage, LG AI Research, SK Telecom, NC AI, and Naver," and the government later ran a supplemental round to add a fourth entrant: Motif Technologies, "a neolab that spun out of Moreh." Motif is tiny by any standard. "Motif was a sub 30 person startup that released their first pre-trained model, which was only 2.6B total parameters, last June," the report says, adding that the team has "raised just $17M and have access to a mere 768 B200s." Even so, per the report, "Motif 3 is 10 points ahead of Upstage and by far the best Korean model."
The upside on the supply side looks distinctly American. Samsung "reportedly plans to build an Nvidia-powered AI factory using more than 50,000 GPUs," and SK Telecom's "announced 2GW DSX AI factory is expected to deploy Vera Rubin systems powered by SK Hynix HBM4." The gigawatt ambition sits oddly next to the state-level US datacenter pushback we've been logging this week, where governors are curbing individual sites at a fraction of Korea's scale. Pennsylvania signed an order curbing 'predatory' data centers yesterday. Against the datacenter number, meanwhile, the government's model-training pot is small: "the total government budget of ~$350M is still a rounding error compared to US labs."
Jensen Huang, quoted in the piece, argues that "Open weight models—AI models that anyone can download, inspect, modify, and run on their own infrastructure—are an important part of that foundation," a framing that fits Korea's insistence on locally-built open models even while it imports American silicon to run them.
Originally reported by newsletter.semianalysis.com
Read the original article →Original headline: SemiAnalysis Details South Korea's $919B Sovereign AI Push, 18.4GW Datacenter Target by 2035