SK Telecom said on April 22 it unveiled the results of its technical cooperation with Nvidia in AI model development at 'Nvidia Nemotron Developer Days Seoul 2026', held in South Korea for the first time on April 21.

At the event, SKT introduced plans for cooperation for a domestic sovereign AI ecosystem, including development of A.X K2, a follow-up model to A.X K1 that was presented through the government's independent AI foundation model project.

The two companies have exchanged expertise across data, infrastructure and the overall training environment since the stage when SKT built its A100 GPU-based supercomputer Titan, unveiled in 2021. They have since continued technical cooperation for SKT's development of its own AI models.

In particular, SKT has adopted the Nvidia Nemotron dataset for training A.X K1. In the process, SKT and Nvidia used Megatron LM, a large-model distributed training framework, and Nvidia NeMo Curator for preparing and refining training data to boost the stability of training for very large AI models.

SKT's model development unit and Nvidia's technical support unit communicate every two weeks to discuss training infrastructure stability, performance improvements and optimisation measures. The two companies shared experience in building very large models using new architectures such as mixture of experts, as well as information related to development infrastructure, laying the groundwork to carry out large-scale training more precisely and stably.

SKT said it will use Nvidia solutions in training A.X K2, the follow-up model currently under development, and continue exchanges needed to advance the companies' technological capabilities. It will also continue active joint research in next-generation foundational technology areas, including multimodal and vision-language models to be used in future models.

Kim Tae-yoon (김태윤), head of SKT's foundation model work, said, "SKT and Nvidia have applied new technologies proactively and developed together through mutual feedback." He added, "Through cooperation, we will contribute together to the development of the domestic AI ecosystem."