In Viet Nam, AI is becoming increasingly integrated into sectors, such as finance, healthcare, manufacturing, education, and public administration. Data is no longer simply stored; it is continuously fed into AI models for learning, analysis, and decision-making. If this process relies on foreign AI platforms, the risks extend beyond sensitive data passing through infrastructure located outside the country's borders. The ability to generate knowledge, develop AI models, and create economic added value also becomes dependent on external providers. Establishing an "AI heart" domestically addresses both challenges: protecting data while creating a foundation for developing AI models that understand the Vietnamese language, culture, laws, and the specific characteristics of Vietnamese data.
    |
 |
|
Viettel operates the first data center in the Asia-Pacific region to integrate NVIDIA's new H200 GPU chips. |
In Viet Nam, the development of sovereign AI is being promoted through AI-Native Cloud platforms, infrastructure designed with AI at the core of their architecture from the outset, rather than simply adding AI capabilities to traditional cloud platforms. A notable example of this approach is the Military Industry-Telecoms Group (Viettel) Cloud AI-Native, an infrastructure platform developed and controlled by Vietnamese engineers. It enables domestic organizations and businesses to build, train, and run intelligent applications entirely within Viet Nam.
The system is housed at the Hoa Lac 2 Data Center in Ha Noi, the first data center in the Asia-Pacific region to integrate NVIDIA H200 GPUs and receive the TCCF Uptime Tier III certification for reliability and continuous operations.
Viettel Cloud AI-Native is equipped with NVIDIA H200 SXM5 GPUs, a powerful generation of AI processors featuring 141 GB of memory per GPU. This enables the platform to run large language models (LLMs) with 70 billion or more parameters without having to split the models into smaller parts.
To fully harness the hardware's processing power, the system incorporates InfiniBand NDR400 high-speed networking, allowing GPUs to communicate directly with one another at speeds of up to 400 Gbps. This eliminates bottlenecks when training AI models across multiple servers in a distributed computing environment.
In addition to its hardware, Viettel Cloud AI-Native comes integrated with NVIDIA AI Enterprise (NVAIE), a software suite that includes tools for model training (NeMo), inference acceleration (TensorRT), and AI model deployment (Triton Inference Server). The entire system is centrally monitored through a single dashboard, giving customers an overall view of operations while providing automatic alerts when abnormalities occur.
Organizations and businesses handling sensitive data can confidently adopt the solution thanks to its multi-tenant architecture, which provides three independent layers of isolation: computing resources, stored data, and internal networks. Multiple organizations can share the same system while keeping their data completely separate, minimizing the risk of unauthorized cross-access. The entire infrastructure is located in Viet Nam and operated by Viettel engineers, ensuring that data does not pass through to foreign servers.
Building an "AI heart" is about more than infrastructure or processing speed. It lays the foundation for Viet Nam to master the entire AI value chain, from infrastructure and data to operations, and to independently determine how AI is applied, controlled, and leveraged, rather than relying on external technologies and rules.
Translated by Tran Hoai