FPT is ready to provide GPU Cloud services powered by NVIDIA HGX B300 through FPT AI Factory, opening up a next-generation AI infrastructure platform for enterprises, organizations, and AI developers.
FPT Corporation recently announced the deployment of new AI infrastructure at its AI Factory using NVIDIA HGX B300 systems, an advanced GPU platform designed for large-scale AI data centers. The new computing infrastructure is designed to support reasoning AI and agentic AI, models capable of making decisions and taking action rather than merely processing data.
FPT AI Factory's new infrastructure uses NVIDIA HGX B300 with eight NVIDIA Blackwell Ultra GPU cards. Each GPU has 288 GB of memory, bringing total capacity to 2.1 TB per node and enabling more efficient deployment of trillion-parameter models, multimodal AI, and long-context applications.
Compared with the previous NVIDIA Blackwell generation, NVIDIA HGX B300 delivers up to 1.5 times higher performance, accelerating both AI training and inference. This allows enterprises to shorten model deployment time, serve more users, and optimize operating costs.
The platform can help reduce AI inference costs by up to 66%, lower training costs by 49%, and improve cost efficiency per token by nearly threefold. In addition to performance, the system is built to enterprise security standards, offers high reliability, and is supported by a team of specialized engineers.
The AI industry is entering a new phase in which the focus is shifting from model training toward AI inference, the process through which AI generates responses and handles user requests in real time. This creates demand for processing large token volumes at high speed, with low latency and optimized costs.
In this context, computing infrastructure must evolve to support the “token economy,” where AI performance is measured by token-generation speed, response time, and cost per token. Integrating NVIDIA HGX B300 enables FPT AI Factory to operate as an “AI Token Factory,” helping organizations strengthen AI inference capabilities with high performance and optimized costs.
FPT is the first provider in Vietnam to put NVIDIA HGX B300 into operation, delivering modern AI infrastructure aligned with international standards for large-scale AI research, development, and application. Previously, FPT AI Factory pioneered GPU Cloud services using NVIDIA HGX H100 and HGX H200 systems, serving more than 18,000 AI developers and engineers from enterprises, research institutions, and educational organizations worldwide.
In the TOP500 ranking announced in June 2025, FPT's AI factories in Japan and Vietnam ranked 36th and 38th, respectively. This achievement placed FPT AI Factory among the world's leading supercomputing infrastructures and established FPT as the No. 1 commercial AI Cloud service provider in Japan using NVIDIA H200 Tensor Core GPU SXM5 superchips.
TOP500 is a prestigious global ranking of high-performance computing supercomputers based on the LINPACK benchmark, which evaluates a system's ability to perform complex numerical calculations per second, measured in FLOPS.