Naver, NVIDIA, and Brookfield announced a $10 billion agreement on July 25 to expand Korea's sovereign AI infrastructure from 55 megawatts to 200 megawatts by 2028. The deal positions Seoul as the third major non-U.S. AI compute hub after Singapore and Abu Dhabi, with Brookfield providing structured capital, NVIDIA supplying GPU clusters, and Naver operating the facility under Korean data sovereignty rules.
The initial 55-megawatt phase began construction in Q1 2024 with 14,336 NVIDIA H100 GPUs and came online in May. The expansion adds three sequential buildouts: 50 megawatts in Q4 2025, another 50 megawatts in Q2 2027, and a final 45 megawatts in Q1 2028. Brookfield structured the financing as a $6.2 billion senior facility with $3.8 billion in convertible preferred equity tied to Naver's cloud revenue multiples. NVIDIA receives hardware payments in tranches plus a 7% equity stake in the infrastructure vehicle. Naver retains operational control and first call on 60% of compute capacity for its HyperCLOVA X foundation model and enterprise clients.
This matters because sovereign AI factories are becoming geopolitical instruments, not just data centers. Korea now joins a short list of nations with over 150 megawatts of dedicated AI compute under national oversight. The structure also signals a template: Brookfield brings patient infrastructure capital, NVIDIA monetizes hardware at scale without balance-sheet risk, and the domestic tech champion locks in strategic compute at cost. For allocators, the secondary effect is pricing pressure on hyperscale cloud margins in Asia-Pacific. Naver can now offer inference and training at 30-40% below AWS Seoul pricing for Korean enterprises, forcing Amazon and Microsoft to either match or cede local market share. The facility also backstops Korea's AI export ambitions — Samsung and LG have already reserved 25 megawatts of capacity for semiconductor design workloads and robotics training.
Operators should track three follow-on events. First, watch for Naver's Q3 2025 earnings call in late October, when management will disclose utilization rates and contracted revenue from the initial phase — anything above 75% utilization suggests the expansion timeline accelerates. Second, monitor Brookfield's infrastructure fund filings in Q1 2026 for replication attempts; if they announce similar deals in Vietnam or Indonesia, the sovereign AI factory becomes a repeatable asset class. Third, Samsung's February 2026 foundry roadmap will clarify whether they co-locate an AI chip design facility onsite, which would validate the compute-plus-fabrication cluster model.
By 2028, Korea will control more sovereign AI compute than the combined capacity of Taiwan, Japan, and Australia. Brookfield now holds infrastructure exposure to the one cost input — GPU clusters at scale — that every AI-native government needs and cannot easily substitute.