SK Telecom is accelerating its push into next-generation AI data centers through a strategic partnership with Arm and South Korean AI chip startup Rebellions, as the industry shifts from training-focused infrastructure to inference-driven efficiency.
The company announced on April 9 that it has signed a memorandum of understanding (MOU) with Arm and Rebellions to collaborate on AI infrastructure innovation. The three companies plan to jointly develop a server solution that combines Arm’s next-generation data center processor, the Arm AGI CPU, with Rebellions’ AI inference chip, RebelCard™, and validate the technology in SK Telecom’s AI data centers.
The partnership reflects a broader shift in the AI landscape. While large-scale model training has traditionally defined competitiveness, inference—the stage where AI models deliver real-time services—is emerging as a key battleground. Because inference workloads run continuously, power efficiency has become directly linked to cost competitiveness.
To address this shift, the industry is moving beyond GPU-centric architectures toward alternatives optimized for inference, particularly neural processing units (NPUs). Hybrid computing architectures that combine CPUs and NPUs are gaining traction, with CPUs handling data processing and system operations, and NPUs dedicated to AI computations—enabling both performance and efficiency gains.
The solution being developed by the three companies follows this approach. Arm’s AGI CPU will manage system control and data processing, while Rebellions’ RebelCard will handle AI inference workloads. The architecture is expected to reduce power consumption and lower overall server operating costs.
Arm AGI CPU marks Arm’s first in-house data center chip and is optimized for AI inference workloads. Rebellions’ RebelCard is designed specifically for large-scale AI inference environments, offering a balance of high performance and energy efficiency.
The companies have already demonstrated the commercial potential of this combination. At the “Arm Everywhere” event in March, they showcased a real-time agentic AI service powered by large language models running on the integrated chip solution.
SK Telecom plans to deploy the solution in its AI data centers to verify performance and stability, while also exploring its use for running its proprietary AI foundation model, A.X K1. The company aims to strengthen its competitiveness in the AI data center business through this integrated approach.
Lee Jae-shin, head of AI business development at SK Telecom, said the company will continue enhancing its AI data center capabilities by combining inference-optimized infrastructure with its own AI models.
Eddie Ramirez, Vice President at Arm, noted that surging demand for AI inference is driving innovation in data center infrastructure, adding that the collaboration represents an important step toward next-generation AI systems.
Oh Jin-wook, Chief Technology Officer of Rebellions, said the company aims to establish NPUs as a core component of AI data centers through high-performance, energy-efficient hardware and software capabilities, adding that the partnership could set a meaningful precedent for the industry.
