Krafton said it will reorganize its operations around AI and invest over 100 billion won (US$69.7 million) to become an “AI-first” company.
The South Korea-based game developer plans to build a GPU cluster for its in-house AI infrastructure, with completion targeted for the second half of 2026.
It will also allocate 30 billion won (US$20.9 million) per year from 2026 to support employees in using AI tools for work.
The company said it will restructure HR and organizational systems to integrate AI into workflows, management, and decision-making.
Krafton highlighted a focus on AI-driven automation, employee training programs, and new internal mobility opportunities.
These initiatives aim to improve productivity, support new product development, and strengthen its position in the global gaming market.
🔗 Source: Krafton
🧠 Food for thought
Implications, context, and why it matters.
Krafton’s GPU spend is modest relative to hyperscale projects and lacks clear ROI (return on investment) benchmarks
- KRW 100 billion (~$75 million USD) buys a GPU stack compared with hyperscale (massive cloud scale) builds in South Korea, where SK Telecom (a major Korean mobile carrier) and AWS plan a 103 MW site with 60,000 GPUs and 20 kW per rack 1.
- The widely cited record Q3 2024 operating profit of $5.1 billion belongs to SK Hynix, not Krafton 2. Krafton gave no view on operating cash flow impact or expected gains from AI automation.
- From 2026, KRW 30 billion per year (~$22.5 million USD) goes to AI tools. The company calls it more than ten times current support, yet set no baseline or targets, so margin or schedule impact stays unclear.
Korean colocation providers can capitalize on immediate demand for GPU-ready infrastructure from mid-sized operators
- This GPU cluster creates demand for data centers running high density liquid cooled colocation services (third-party facilities that host customer servers), especially around Seoul and Pangyo 1.
- Digital Realty runs the ICN10 site with advanced cooling for KakaoBank’s AI lab (KakaoBank is a South Korean digital only bank) 2, and cold plate cooling (a method that circulates coolant through plates attached to chips) holds a 48% revenue share 3. Industry sources cite 30 to 50% energy savings from liquid cooling 4, which makes GPU ready colocation attractive to mid sized gaming or fintech teams seeking AI without hyperscale budgets.
Recent Krafton developments
