KaKao

- Establishing a stable operating environment for the latest B200 GPUs at its own data center

- Completed 255 nodes—more than 4 times the original 64-node annual target—achieving 84% of total deployment ahead of schedule

- Beta service to begin on January 2 next year, providing advanced computing resources to selected industry–academe–research projects

 

[December 29, 2025] Kakao (CEO Shina Chung) announced that the company is making smooth progress with the deployment of next-generation GPU infrastructure after being selected as the final operator for the government-led Graphics Processing Unit (GPU) Securing Project, marking a critical milestone in supporting Korea’s AI research and development ecosystem.

 

The national initiative aims to strengthen Korea’s position as a global AI leader by expanding access to critical GPU infrastructure for the private sector. Kakao was selected as the project’s final operator in August. Through the program, Kakao will secure and deploy a total of 2,424 B200 GPUs and operate the infrastructure on consignment for 5 years to support domestic AI research and development.

 

Kakao is deploying a large-scale GPU cluster at a stable rate at its Kakao Data Center in Ansan, Gyeonggi-do. The company accelerated its deployment timeline by leveraging its in-house data center capabilities and extensive experience in building and operating GPU clusters. It has already completed infrastructure for 255 nodes (2,040 GPUs), or approximately 84% of the total allocation. This figure exceeds the company’s original year-end deployment target of 64 nodes by more than 4 times.

 

The early achievement was made possible by the advanced infrastructure readiness of Kakao Data Center Ansan. Kakao implemented comprehensive project management across the entire process from GPU procurement and installation to operational readiness. Close collaboration with suppliers enabled the early acquisition of key equipment, with pre-deployment technical validation (PoC) helping identify and minimize potential risks; thus further accelerating the rollout schedule.

 

The Ansan data center is also equipped with stable power supply systems and cooling infrastructure essential for operating high-density servers. In particular, Kakao implemented a Hot Aisle Containment System that isolates hot air and channels it directly to cooling units, significantly improving thermal efficiency in high-performance GPU servers.

 

Aside from hardware infrastructure, Kakao also provides a software environment that enables users to focus on AI model development. Through an integrated platform linked to the national AI computing resource support portal, users can seamlessly access Kakao Cloud operated by Kakao Enterprise. The platform offers Kubeflow, Kakao Cloud’s AI solution for building and running machine learning workflows in a cloud-native environment with enhanced speed and accessibility. Kubeflow supports the full machine learning lifecycle—including model development, training, deployment, and inference—within a Kubernetes-based environment, enabling researchers to automate workflows and manage cloud resources efficiently.

 

Kakao is currently conducting network and performance tests on the 255 completed nodes. Starting January 2 next year, the company will provide advanced computing resources to industry, academic, and research projects selected through a beta service program organized by the Ministry of Science and ICT and the National IT Industry Promotion Agency.

 

“Building and operating large-scale GPU infrastructure in a stable manner is a key component of AI competitiveness. Based on Kakao’s data center and cloud capabilities, we will continue to provide a reliable, efficient AI development environment and contribute to the growth of Korea’s AI ecosystem,” said Se-woong Kim, Performance Lead of Kakao AI Synergy. (E.O.D.)