Korea’s AI buildout just moved past the blueprint stage. Now comes the expensive part: making all that capacity useful. Before today’s development, some context: NAVER, NVIDIA, and Brookfield are building Korea’s sovereign AI infrastructure at NAVER’s GAK Sejong hyperscale data center. The partners expanded an initial 55 MW plan to 200 MW by 2028. NVIDIA plans a $1 billion investment, Brookfield up to $9 billion, and NAVER has framed the broader ambition as up to 1 GW. This is AI Daily Briefing. Korea has three major compute builds moving at once. We’re asking who gets affordable access when the doors finally open. For updates on this story — NAVER-NVIDIA Korea AI factory expansion — tap follow so the next episode lands in your feed. Here's The Asia Business Daily:
The key infrastructure that will determine the competitiveness of South Korea's artificial intelligence (AI) industry, the National AI Computing Center (KOACC), was launched with a spectacular fireworks show in Haenam, Jeonnam. The government and private sector are investing a total of 2.4 trillion won in capital, with plans to establish Haenam Solar City as a global AI hub.
Haenam has gone from a 2.4 trillion won ambition to an actual construction site. KOACC’s first design is 40 megawatts, expandable to 80, giving founders a 2028 date they can finally plan workloads around. Fireworks at a data-center groundbreaking is certainly a choice. But having Samsung SDS, Naver, and the science ministry on that stage says Seoul sees this as national industrial policy, not a regional server farm. Putting it beside the Gwangju semiconductor complex makes sense. Keeping compute close to chips is the strategy. Now somebody needs to publish what an inference job in Haenam will cost versus AWS in 2028. Korea is building a real stack here—power, compute, semiconductor adjacency, and public capital. The hard part comes after the ribbon-cutting: keeping that capacity busy enough to justify 2.4 trillion won. Edaily, with Lee So-Hyun:
The detailed project structure for the gigawatt (GW)-class AI infrastructure plan, first unveiled last June, has now been revealed. Naver plans to secure a stable supply of graphics processing units (GPUs) through its equity partnership with NVIDIA and reduce the financial burden associated with large-scale facility investments by leveraging Brookfield’s investment structure.
Following up on NAVER’s Korea AI factory buildout: the first 200-megawatt phase now has a financing and operating structure. Honestly, this is the first deal this week where somebody seems to have made the economics work. Nvidia puts in $1 billion of equity. Brookfield can put up to $9 billion into an SPV that owns the GPUs and facility, while Naver’s operating subsidiary pays by usage. A regulatory filing is a lot more useful than a heroic render and a gigawatt headline. It transfers risk in a sensible direction: Brookfield owns the hard assets, Naver sells compute, and Nvidia has skin in the supply relationship. But 200 megawatts is still a giant tab—show founders what a million inference tokens costs on this thing, then we can celebrate. Korea’s buildout is no longer a one-national-center story. We just covered Haenam; now Naver has a documented 200-megawatt commercial structure. Capacity is coming from several directions at once. From Huda Ata at Gulf News:
The expansion will see Aleria deploy up to 16,000 Nvidia Blackwell Ultra graphics processing units (GPUs) in the United States, alongside 28 racks of next-generation Nvidia DGX Vera Rubin NVL72 systems in the UAE, marking one of the first deployments of the platform in the region.
Aleria gets credit for saying what most of these announcements dodge: this is already running. Eight thousand six hundred forty Blackwell Ultra GPUs are in a 25-megawatt U.S. facility, scaling toward 16,000. Those are deployment details you can test. And 28 DGX Vera Rubin NVL72 racks in the UAE, among the region’s first. “Sovereign” means the data and facility sit under local control; the compute roadmap still arrives on Nvidia’s timetable. Right. Aleria has a serious stack: DDN storage paired with Nvidia accelerated compute for sensitive government workloads. But when a founder asks what an inference job costs there versus a hyperscaler, the sovereign pitch has to meet production. We just had Korea’s financing blueprint. Aleria is a useful parallel because it has named hardware and live sites on two continents. Capacity projections are plentiful. Operational racks are a better kind of fact. From Paige Gross at CIO Dive:
Enterprise spending on cloud infrastructure services reached $143 billion in the second quarter of 2026, up from $100 billion in the previous quarter, according to Synergy Research Group data. The year-over-year growth rate notched the 11th consecutive growth quarter jumping 43% — the highest growth rate in the last eight years, the market analyst group found.
Synergy puts Q2 cloud-infrastructure spend at $143 billion, up 43% year over year—the fastest growth rate in eight years. That puts a demand curve behind Amazon’s AI spending binge, not just an earnings-call adjective. AWS still has 28% of this market, ahead of Azure at 20 and Google at 15. Fine. But businesses deploying agents need usable throughput at a price that survives a month-end bill, not another glossy capacity chart. Gen-AI cloud services are growing 165%, according to Synergy. That $500 billion annual cloud-spend figure explains why every board and government ministry suddenly has a data-center plan. Demand is clearly there. The next test is whether AWS and everyone building alongside it can turn that demand into cheaper inference, rather than just pricier waiting rooms. Mingtiandi writes:
The board of LG Uplus on 29 July approved KRW 1.3 trillion ($895.6 million) in funding for the second phase of its Paju AI Data Centre, according to a filing with South Korea’s Financial Supervisory Service, with the investment representing a follow up to the KRW 615.6 billion which LG committed to the first phase in May 2025.
LG Uplus’ board approved another $895.6 million for Paju on July 29—200 megawatts, liquid cooling, eventually 100,000-plus servers. After Haenam and Naver, Korea’s buildout has three separate institutions moving capital at once. It’s moved into coordinated spending. LG’s phase-one check was KRW 615.6 billion; this new KRW 1.3 trillion commitment is more than double that. So yes, Paju is getting real. But a 200-megawatt campus 31 kilometers north of Seoul still needs to tell founders what an inference job costs there versus AWS or Naver in 2028. The timing is revealing: LG approved the money the same day Seoul launched the AIDC Alliance, targeting 18.4 gigawatts by 2035. The alliance includes Samsung SDS and SK, plus Kakao and FuriosaAI. It’s a corporate coalition with government scale behind it, and LG Electronics is supplying the cooling inside its own group. Eighteen-point-four gigawatts is a handsome target. Paju’s 200 megawatts and an actual board authorization are what I can put on a spreadsheet. Now show the utilization and the pricing. For more on AI’s impact in government and defense, try Anthropic Pentagon Watch, a daily briefing on Anthropic’s fight with the DoD over Claude, military AI use, autonomous weapons, and procurement blacklisting. Find it wherever you get podcasts.
We’re watching the National AI Computing Center in Haenam, which is planned to begin operations in 2028.
Links to every story are in the show notes, so check out the pieces that caught your attention. That’s AI Daily Briefing for today. This is a Lantern Podcast.