SK Telecom (SKT) revealed plans to build what it described as an AI infrastructure superhighway, starting with the launch of a cloud-based GPU-as-a-service (GPUaaS) through a partnership with US-based Lambda.

In a statement, the operator explained its first GPUaaS will use Nvidia H100 Tensor Core’s GPUs and in March 2025 would tap the chipmaker’s H200 Tensor Core GPU. SKT added it aims to enable companies to develop AI services “easily and at a lower cost”, without needing to purchase their own GPUs. 

The company will convert a data centre in the Seoul metropolitan area into AI-powered facility to address what it calls a domestic GPU shortage, and open a testbed in Pangyo in December.

The announcement was made at a two-day AI event, where CEO Ryu Young-sang outlined the operator’s AI strategy.

Ryu said it plans to expand globally by collaborating with partners in South Korea and abroad, providing AI data centre services. He added competition in telecoms infrastructure had been all about connectivity, “but now the paradigm of network evolution should be changed. The upcoming 6G will evolve into a next-generation AI infrastructure where communication and AI are integrated.”

SKT will start with hyperscale data centres requiring more than 100MW in the domestic market, with plans to expand the scale to gigawatts, with the objective to become a hub for the Asia-Pacific region.

The operator also plans to introduce edge AI, with the aim to narrow the gap between AI data centres and on-device AI, using the nationwide communication infrastructure.

It noted edge AI combines mobile networks and AI computing, which reduces latency, increases security and improves privacy compared with large-scale data centres.