SK Telecom, Rebellions expand Korean AI chip infrastructure, in bid for sovereignty over NVIDIA chips

Key Takeaways:

Nvidia Controls AI Model Training: Driven by its high-performance GPUs and CUDA software ecosystem, Nvidia powers 92% of the \~170 sovereign large language models (LLMs) developed across 80+ countries.

Shift Toward AI Inference: As the AI market transitions from model training to running daily services (inference), data center priorities are shifting toward cost efficiency and energy consumption.

South Korea's Domestic Alternative: SK Telecom and AI chipmaker Rebellions are building an inference infrastructure using locally developed semiconductors to reduce reliance on Nvidia for service operations.

Workload-Specific Infrastructure: The market is evolving into a hybrid model where companies continue relying on Nvidia GPUs for training hyperscale models, while deploying specialized alternative chips for cost-effective inference.

https://www.upi.com/Top\_News/World-News/2026/08/07/skorea/9321786142003/