5nm AI accelerator cards and servers with dataflow architecture for inference
Rebellions manufactures 5nm AI accelerator hardware optimized for HBM and chiplet layouts, paired with a full-stack software layer supporting PyTorch 2.x and vLLM. The company is in mass production and already shipping thousands of ATOM cards to telecom operators. Hiring is heavily skewed toward senior engineering (9 of 15 open roles), with active projects spanning driver development, network scaling POCs, and production system code—indicating a transition from R&D into manufacturing operations and supply-chain scaling.
Rebellions is a Seoul-based AI chip startup founded in 2020, designing and manufacturing 5nm AI accelerator cards, servers, and rack-scale systems for datacenters. The product includes custom ASIC hardware optimized for memory bandwidth and chiplet interconnect, plus a proprietary software stack that integrates with existing AI frameworks (PyTorch, vLLM). The company has begun mass production and is already deployed in commercial AI services at telecom operators. Current operational focus spans hardware bring-up, device driver support, supply-chain lead-time optimization, and early-stage partnership sales.
Core languages: C, C++, Python. System-level: Linux, RTOS, KVM, SR-IOV, PCIe, IOMMU. Networking: Ethernet, RDMA, InfiniBand, TCP/IP. AI frameworks: PyTorch, TensorFlow, NCCL, MPI. Adopting vLLM.
Mass-producing 5nm accelerator cards (ATOM), scaling network architecture via POCs, driver development for RTOS, AI/ML task scheduling software, and supply-chain optimization to reduce lead times.
Other companies in the same industry, closest in size