SKT partners with Arm, Rebellions on hybrid CPU-NPU servers

An Arm AGI central processing unit (CPU) / Courtesy of Arm

SK Telecom has teamed up with semiconductor design giant Arm and artificial intelligence (AI) accelerator startup Rebellions to develop a hybrid central processing unit (CPU)-neural processing unit (NPU) solution to empower next-generation AI infrastructure.

SK Telecom said Friday it signed a strategic memorandum of understanding with the two companies.

Under the agreement, the three parties will jointly develop an AI inference server solution that integrates Arm’s new Arm AGI CPU with Rebellions’ AI accelerator chip RebelCard, which is slated for release in the third quarter. The solution will be tested and verified in SK Telecom’s AI data centers.

Arm AGI CPU, which marks Arm’s first architecture for data center processors, is optimized for high-density AI inference workloads and large-scale AI deployment. Rebellions’ RebelCard is specialized for large-scale AI inference.

The partnership comes as the AI industry undergoes a paradigm shift from training AI models to powering real-world services, driving demand for data center architectures that balance performance and power efficiency for round-the-clock inference operations.

Combining NPUs with CPUs creates a heterogeneous computing architecture where the CPU handles general-purpose computing, such as data processing and system operation, and the NPU performs dedicated AI inference, enhancing overall performance and energy efficiency while reducing operating cost.

The company plans to install servers equipped with this solution at its AI data centers to validate performance and stability, while it is also reviewing how to operate its proprietary AI foundation model, A.X K1, on these servers.

“By offering a full package that combines infrastructure optimized for inference with our sovereign AI foundation model A.X K1, we will further enhance the competitiveness of our AI data centers,” Lee Jae-shin, SK Telecom’s head of AI business development, said.

At Arm’s Arm Everywhere event in March, Arm and Rebellions showcased a live demonstration of an agentic AI service based on OpenAI’s GPT OSS 120B running on the combined chips, highlighting the potential applicability of this architecture for large-scale data center workloads.

“The rapid growth of AI inference is driving demand for new data center infrastructure designed for large-scale deployments. Partners like SK Telecom and Rebellions played an important role in helping us build the Arm AGI CPU and modernize AI inference infrastructure,” Eddie Ramirez, Arm’s vice president of cloud AI business unit, said.