KT Group invests 15 billion KRW in AI company 'More' to accelerate 'AI Full Stack' business

KT and KT Cloud Invest 10 Billion and 5 Billion Respectively
Leading Global Competitiveness with Domestic AI Technology

On the 23rd, KT Group announced that it has made a 15 billion KRW investment in 'More', an AI infrastructure software specialized company, to accelerate its 'AI full-stack' business.

This is the second investment in More following the one in 2021. KT is investing 10 billion KRW, and KT Cloud is investing 5 billion KRW.

KT has been executing an AI full-stack strategy that encompasses domestically pure technology-based software and hardware together with major tech companies representing South Korea's AI semiconductor ecosystem, such as KT Cloud, More, and Rebellions.

AI full-stack refers to products and services covering the entire process from AI infrastructure such as AI semiconductors and cloud, which form the foundation of AI business, to AI application services provided to customers. Companies possessing an AI full-stack environment are rare worldwide.

In building computing infrastructure necessary for AI service development, NVIDIA's GPU market share approaches 80% globally. The high dependence on NVIDIA is because most AI services and solutions are developed based on NVIDIA's software 'CUDA (Compute Unified Device Architecture)'. CUDA is software required to utilize GPUs (graphics processing units) for AI computation and development, and without support, there are limitations in using GPUs for AI computation.

KT is collaborating with More, Rebellions, and others to overcome dependence on foreign GPUs, secure mid- to long-term AI capabilities, and strengthen national AI competitiveness. In particular, More is an AI infrastructure software specialized startup providing ▲ AI infrastructure software stack ▲ AI cluster and cloud solutions. By applying the AI infrastructure software stack first, it provides infrastructure software compatible with existing CUDA, offering the same development environment to resolve dependence and dependency on foreign GPUs. Ultimately, chipsets developed by domestic AI semiconductor fabless companies like Rebellions can also be easily applied to AI development environments.

Additionally, the AI cluster and cloud solutions solve the difficulties of implementing large AI models through automatic parallelization of compilers and improve cloud infrastructure utilization through GPU virtualization and dynamic allocation. Programs written for a single GPU can be automatically parallelized and executed on multiple GPUs, making it easier to implement large-scale artificial neural network models like GPT-3. When GPU computation is executed, a virtual GPU is allocated, and once the program ends, the allocation is released, allowing flexible operation of computing infrastructure resources as needed.

KT combined More's technology with KT Cloud infrastructure to launch the world's first pay-as-you-go GPU service, 'Hyperscale AI Computing (HAC)', at the end of 2020. HAC has succeeded in applying its own AI framework that supports CUDA and is supporting the provision of the same development environment for GPUs from semiconductor companies other than NVIDIA.

A major advantage of HAC is that it is a pay-as-you-go service based on actual usage, providing high-cost GPU infrastructure through dynamic allocation. AI service companies or AI developers can allocate and use GPU resources only as much as they want at the desired time and return them afterward. In actual development processes, GPUs are not needed throughout the entire development period but are used only for data analysis, model training, etc. Using this service allows GPUs to be used only for the necessary period, significantly reducing development costs.

A KT official said, "This investment is expected to be an important milestone in securing national AI competitiveness based on domestic technology in a situation where the domestic AI infrastructure market largely depends on foreign solutions," adding, "We will concretize the common goal and execution plan of global AI full-stack expansion based on the cooperative relationship between the two companies."

This content was produced with the assistance of AI translation services.