Naver Cloud and Vietnam's FPT signed a memorandum of understanding at Naver 1784 in Seongnam to co-develop AI factory and GPU-as-a-service offerings, giving enterprise customers in Korea, Japan, and Vietnam a single path to train and run HyperCLOVA X models without building their own clusters.
Chief executive Kim Yu-won said the deal combines Naver's hyperscale language-model stack with FPT's NVIDIA-backed AI factory deployments, targeting conglomerates that want sovereign hosting in Asia rather than defaulting to U.S. hyperscalers.
What the factories will sell
Under the MOU, engineers from both companies will joint-tune inference stacks, optimize GPU scheduling, and prototype agent workflows for manufacturing and finance clients. FPT already markets an AI factory platform; Naver will expose HyperCLOVA X through that channel while retaining model governance in Korea.
The partnership follows Naver Cloud's selection to lead a government cybersecurity-AI consortium and its defense-AI day presentations in September. Regulators want domestic clouds to prove they can host sensitive workloads as automated welfare and financial investigations lean harder on machine learning.
Competitive context
Korean enterprises have complained that GPU reservations remain tight even after additional datacenter builds in Pangyo. By linking to FPT's Vietnam footprint, Naver gains burst capacity without immediate capex in Korea, while FPT gains a flagship model partner beyond its services roots.
Rivals such as KT Cloud and Kakao's cloud unit will watch whether the MOU includes exclusivity on certain verticals. Both companies said Tuesday they are pursuing similar Southeast Asian reseller deals, suggesting the region is becoming a battleground for inference hosting rather than raw training.
Customer timeline
Executives declined to name launch customers but said pilot integrations would start this quarter with FPT's existing factory tenants. For Korean regulators, the test is whether data residency clauses survive when workloads spill into Vietnamese racks—a question likely to surface in upcoming cloud security audits.
Sovereignty and compliance
Personal information protection lawyers said cross-border inference must still honor Korea's data-localization guidelines for public-sector clients. Naver Cloud argued that training can remain in Seongnam while inference bursts in Vietnam, but contracts will need explicit purpose limitation clauses. Financial supervisors have not ruled on whether Vietnamese inference violates outsourcing rules for banks.
Startups welcomed cheaper GPU hours, predicting fine-tuned models for logistics and insurance claims. Critics noted concentration risk: if two vendors dominate APAC AI factories, a outage in either country could stall dozens of pilots. Naver pledged dual-region failover in its service level objectives, details to come at a November developer day.
Developer ecosystem
Naver plans workshops for ISVs on fine-tuning HyperCLOVA X with LoRA adapters, a pitch aimed at public-sector chatbots that must cite sources. FPT will mirror those sessions in Ho Chi Minh City, hoping to recruit Vietnamese fintech engineers who already work on NVIDIA stacks.