South Korean AI semiconductor startup Rebellions is partnering with Japanese neo-cloud company ai& to supply more than 100 RebelRack neural processing unit (NPU) units to ai&’s Tokyo data center. RebelRack combines servers equipped with Rebellions’ in-house Rebel100 NPU into rack systems designed for energy-efficient AI inference. Japanese enterprises, public institutions and developers will be able to access inference services through the deployed hardware without changing their existing development environments. ai& uses a heterogeneous infrastructure model that combines computing resources according to workload requirements, while both companies support open-source software and major frameworks widely used in GPU environments. The companies said this compatibility could allow commercial services to be deployed from the first day of delivery without large-scale software redevelopment. ai& has secured more than $2 billion in infrastructure investment, plans to operate five data centers by the end of this year and aims to expand the Tokyo facility’s total capacity to 40 megawatts by the end of next year. The initial deployment is expected to scale beyond 100 units. The project is also positioned as an infrastructure example supporting Japan’s sovereign AI strategy and as a commercial test of NPU-based inference infrastructure in a market dominated by NVIDIA GPUs. Rebellions CEO Park Sung-hyun said the partnership would give the company direct access to Japanese customers and help expand its global inference infrastructure business.