SK Telecom, Arm, and Rebellions Partner on AI Inference Server to Boost Energy Efficiency

SK Telecom, Arm, and Rebellions have signed an MOU to co-develop a new AI server that combines Arm's AGI CPU with Rebellions' next-generation NPU.

This move targets the growing AI inference market, where energy efficiency and low total cost of ownership (TCO) are becoming more critical than raw training power.

The partnership is a culmination of recent events, including Arm's new data center CPU launch and significant government investment in Rebellions, signaling a strategic shift towards specialized, power-efficient AI hardware.

017670.KS

SK Telecom, Arm, and Rebellions have announced a major partnership to build a next-generation AI inference server.

This collaboration directly addresses a critical shift in the AI industry. The focus is moving from the energy-intensive 'training' phase of AI models to the 'inference' phase, where models run 24/7 to serve users. According to McKinsey, inference could make up over half of all AI computing demand by 2030. This constant operation makes power efficiency and Total Cost of Ownership (TCO) the most important metrics. The International Energy Agency (IEA) projects that data center electricity use will more than double by 2030, largely driven by AI, making energy-saving solutions essential.

This partnership didn't happen in a vacuum; it's the result of several key developments. First, Arm recently launched its own data center processor, the 'AGI CPU,' moving beyond just licensing its designs. This new chip provides a powerful and efficient foundation for the server. Second, Rebellions, a Korean AI chip startup, secured a major KRW 250 billion investment from Korea's National Growth Fund, ensuring it has the resources to mass-produce its next-generation NPU. Third, SK Telecom has been preparing for this shift by signing deals for modular data centers and advanced CXL memory technology, building an ecosystem ready for this new type of hardware.

By combining these pieces, the companies are creating a 'heterogeneous computing' stack. This means using the right tool for the right job: Arm's CPU for general tasks and Rebellions' specialized NPU for AI inference. This approach is designed to be far more power-efficient than using general-purpose GPUs for everything. An NPU-based server could save approximately 24.5 MWh of energy per year compared to a GPU-based one, a significant reduction in operational costs.

Ultimately, this three-way alliance is a strategic response to the evolving demands of the AI market. It brings together a major telecom operator (the user), a leading CPU platform (Arm), and a specialized AI accelerator company (Rebellions) to create a solution optimized for the future of AI—one that is both powerful and sustainable.

NPU (Neural Processing Unit): A specialized processor designed to accelerate machine learning and AI tasks, making it more efficient than a general-purpose CPU or GPU for these specific workloads.
TCO (Total Cost of Ownership): The complete cost of a hardware or software asset, including the initial purchase price, operational costs like electricity and maintenance, and other related expenses over its lifetime.
Heterogeneous Computing: A system that uses more than one kind of processor or core to improve performance or energy efficiency. In this case, it combines a CPU and an NPU.

SK Telecom, Arm, and Rebellions Partner on AI Inference Server to Boost Energy Efficiency

SK Telecom, Arm, and Rebellions have signed an MOU to co-develop a new AI server that combines Arm's AGI CPU with Rebellions' next-generation NPU.

This move targets the growing AI inference market, where energy efficiency and low total cost of ownership (TCO) are becoming more critical than raw training power.

017670.KS

SK Telecom, Arm, and Rebellions have announced a major partnership to build a next-generation AI inference server.

NPU (Neural Processing Unit): A specialized processor designed to accelerate machine learning and AI tasks, making it more efficient than a general-purpose CPU or GPU for these specific workloads.
TCO (Total Cost of Ownership): The complete cost of a hardware or software asset, including the initial purchase price, operational costs like electricity and maintenance, and other related expenses over its lifetime.
Heterogeneous Computing: A system that uses more than one kind of processor or core to improve performance or energy efficiency. In this case, it combines a CPU and an NPU.