Naver's Nvidia-backed AI factory faces crucial test after US$9 billion deal

This $9B Brookfield-Naver deal is not an infrastructure story—it is the structural fracture of AI compute markets from hyperscaler monopoly into regionalized GPU rental ecosystems. Brookfield solves capex; the kill zone is opex. GPU depreciation cycles of 3-4 years mean Naver must lock anchor tenants within 18 months or replicate CoreWeave's 2024 cash-flow crisis. The technical ripple extends further than headlines suggest: roughly 25-30% of the $9B flows into HBM3E (SK Hynix, Samsung) and InfiniBand switching silicon, directly pulling advanced packaging capacity in Taiwan, China. If liquid-cooling PUE stays above 1.2, Naver's per-FLOP cost runs 15-20% above Texas or Ireland peers. On compliance, Brookfield's Canadian capital structure sidesteps US export controls, but Korea's data-residency mandates effectively cap Naver's addressable market to APAC, blunting its pitch to OpenAI's global inference workloads. Competitively, Samsung's HBM-plus-proprietary-datacenter vertical integration and SK Telecom's Azure Korea node are eroding the same whitespace. Naver's only moat—local latency—is real but narrow in an inference-era market. 12-24 month outlook: H200/B200 volume ramp will compress GPU rental pricing 30-40%. AI factories without anchor tenants enter a cull. Naver's survival hinges on binding Korea's sovereign-AI procurement pipeline; absent that policy tailwind, the $9B asset base faces material impairment by 2026.