FPT and NAVER Cloud sign an MOU to offer GPU rental across Asia-Pacific

An infrastructure lead in South Korea, Japan or Vietnam, the three markets NAVER Cloud named for GPU rental ^[3], cannot set this MOU beside a hardware quote yet. The partners have not disclosed data residency arrangements, service-level commitments, pricing or GPU capacity for the joint offerings ^[6].

The pitch is that enterprises rent compute for training and inference and skip maintaining their own GPU infrastructure ^[12]. The thing done so far is a memorandum that combines FPT's existing computing infrastructure with NAVER Cloud's AI models and engineering expertise ^[2]. That infrastructure is already running: FPT operates NVIDIA-powered AI Factories in Vietnam and Japan ^[7]. DataCenterDynamics lists H100s at 46.56 petaflops Linpack for the Vietnam site ^[8] and H200s at 49.85 petaflops for the Japan site ^[9], or 96.41 petaflops between them ^[15]. Those are benchmark results for machines FPT already operates. FPT opened its Fornix HCM02 data center in Ho Chi Minh City in August 2025, bringing its Hanoi and Ho Chi Minh City facilities to four ^[13].

Here's what teams tell themselves: enterprises across the region are waiting to rent GPUs. Here's what the one adoption figure in the coverage shows. An AWS-commissioned study estimates that 26% of Vietnamese businesses use AI, and most of those adopters are still experimenting ^[17]. We think experimenters are the likeliest renters, because they have no steady load to justify buying hardware, and also the buyers least able to commit to a long contract. FPT says its platform serves more than 18,000 developers and over 2,000 enterprises worldwide ^[16]. Those figures describe the existing platform and do not count customers from the new agreement ^[16].

NAVER Cloud CEO Yuwon Kim said "we will go beyond Korea and Vietnam to address customers' diverse AI needs across the APAC market" ^[10]. Le Hong Viet, CEO of FPT Smart Cloud, said FPT looks forward to working with NAVER Cloud "to bring AI models closer to enterprise customers" ^[11]. NAVER lists six cloud regions in South Korea and one each in Japan, Germany, Singapore and the US West Coast, with Vietnam, Taiwan, Thailand and the US East Coast planned ^[18].

To apply this tomorrow, sort each workload on two axes: whether it can wait for published terms, and whether its data must stay in one country.

- Can wait, data can move: track the MOU and quote existing capacity in the meantime. - Can wait, data must stay in one country: wait for the partners to name sites, because only the sites decide whether they cover your country. - Cannot wait, data can move: ask FPT about the Vietnam and Japan AI Factories that already run ^[7], and judge the MOU later. - Cannot wait, data must stay in one country: unless that country is Vietnam or Japan, the sources list no FPT AI Factory there ^[7], so price hardware or another provider.