TLDR: Arm predicts that specialized low-power CPUs will be a crucial catalyst for the expansion of AI data centers in South Korea, emphasizing the critical need for energy efficiency in the burgeoning AI era. This strategy involves collaboration with Korean firms to build an optimized AI infrastructure.
Arm, a global leader in semiconductor intellectual property (IP), is projecting that customized low-power CPUs will be instrumental in driving South Korea’s shift towards advanced AI data centers. This forecast underscores a growing industry focus on energy efficiency as artificial intelligence workloads demand increasingly significant power consumption. The company is actively working with Korean firms to establish an AI infrastructure designed for minimal power usage.
Mohamed Awad, Senior Vice President at Arm, highlighted the urgency of this transition during a keynote speech at the ‘SMARTCLOUD SHOW 2025’ in Seoul on August 27, 2025. His address, titled ‘The infinite possibilities of AI: Redefining infrastructure for a new era,’ emphasized that Arm’s technology is at the forefront of industries requiring AI, including mobile, automotive, IoT, and data centers. Awad pointed out the immense power demands of AI training, citing that training OpenAI’s GPT-4 consumed 1 million gigabytes (GB) of power, equivalent to approximately 200 million songs. He stressed the imperative to improve ‘performance per watt’—a key power efficiency metric—to mitigate the high energy costs associated with AI.
This strategic direction aligns with broader industry trends where major players like NVIDIA, Amazon, and Google are also investing heavily in technologies that enhance performance while minimizing power consumption in data centers. Arm’s long-standing partnership with companies like Nvidia is a testament to this, with Nvidia utilizing Arm-based CPUs such as Grace and its successor, Vera, for data center applications. These Arm-based CPUs are designed to optimize GPU performance and offer superior power efficiency compared to traditional x86-based CPUs, which often consume more power per unit of performance.
Also Read:
- Nvidia Invests $5 Billion in Intel, Forging Strategic AI and PC Partnership
- Cloud Platforms Emerge as the Indispensable Foundation for AI Data Management
The move towards customized, low-power Arm CPUs is particularly relevant in the context of AI data centers, where soaring electricity expenses represent a significant challenge. While GPUs excel at parallel processing for AI, their combined use with high-performance CPUs can drastically increase per-rack power density and thermal management costs. Arm’s approach aims to address these issues by providing a more power-efficient foundation for AI computation, thereby supporting the sustainable growth of AI infrastructure in South Korea and globally.
