Naver Cloud ramps up use of AMD EPYC, citing high Intel CPU costs

Naver Cloud's 35,000-server AMD deployment is a public rejection of Intel's pricing architecture, not a mere hardware swap. The trigger is arithmetic: when Xeon high-core SKUs breach the TCO ceiling, cloud procurement logic flips from performance-first to cost-per-core-first. Technical cascade: the 144-core Zen 5 rollout forces a complete re-architecture of NUMA topology, memory channel ratios, and PCIe lane allocation. But Naver's software stack is already 18 months ahead in AMD optimization—a migration moat Intel cannot replicate on a quarterly cycle. Market dynamics: Intel's most probable counter is a Korea-specific Xeon 6 channel discount. Yet this directly contradicts its IDM 2.0 playbook, where 18A foundry subsidies are meant to protect CPU ASP globally. AMD, meanwhile, locks in a critical APAC public-cloud reference, tilting negotiation leverage for Line, Kakao, and Southeast Asian cloud operators in its favor. The 12-24 month tail: the real signal is structural. Once a cloud operator crosses the 30,000-server threshold, CPU vendor bargaining power degrades from technology lock-in to price competition. If Intel cannot compress high-core pricing to within 1.3x of EPYC by end-2026, its data center share erosion becomes irreversible.