SK Telecom has forged a strategic partnership with Arm and the innovative Korean AI chip startup Rebellion to spearhead the development of advanced AI data center infrastructure. This collaboration marks a significant response to the industry’s evolving focus, shifting from demanding AI model training towards efficient AI inference processing, as the South Korean mobile carrier announced Friday.
This pivotal partnership, formalized on Thursday, aims to create a groundbreaking server solution. It synergistically integrates Arm’s new data center processor, the Arm AGI CPU, with Rebellion’s highly anticipated AI chip, the RebelCard, which is scheduled for a third-quarter release.
SKT emphasized that market demand within the AI sector is increasingly moving away from training large language models and gravitating towards their practical deployment in real-world applications.
Crucially, AI inference workloads demand continuous operation, making power efficiency a paramount factor in controlling operating costs for AI data centers. To address this, the collaborating companies are strategically leveraging Rebellion’s specialized neural processing units (NPUs), which are purpose-built for highly efficient AI inference processing, thereby aiming to significantly reduce overall power consumption.
Arm’s AGI CPU represents its inaugural in-house data center chip, meticulously engineered for optimal AI inference. Complementing this, Rebellion’s RebelCard is expertly designed to deliver exceptional, high-efficiency performance across vast, large-scale inference workloads.
The companies successfully demonstrated their combined solution at the “Arm Everywhere” event in San Francisco in March. This demonstration featured a real-time agentic AI service powered by a 120-billion-parameter GPT model, powerfully underscoring its immediate commercial viability.
SKT has outlined plans to deploy this innovative AI solution within its own AI data centers to rigorously test its performance and stability. Furthermore, the company is actively exploring the integration of its proprietary foundation model, A.X K1, into this system, aligning with its comprehensive strategy to deliver seamlessly integrated AI infrastructure and software offerings.
“By combining inference-optimized infrastructure with our proprietary foundation model, we aim to strengthen our competitiveness in AI data centers,” stated Lee Jae-shin, head of AI business development at SKT.
Arm Vice President Eddie Ramirez commented that the rapid proliferation of AI inference is the primary driver behind the demand for new data center architectures specifically optimized for large-scale deployment.
Rebellion CTO Oh Jin-wook highlighted that this partnership unites complementary strengths across both hardware and software, setting a new precedent for the construction of specialized AI infrastructure.
