Rebellions said on Sept. 21 that its neural processing units (NPUs), including Atom Max, are being used for SK Telecom AI models powering services such as A. voice synthesis, AI customer centers and Scam Vanguard, expanding commercial deployments since its NPU was introduced for A. call summaries in December 2025. SK Telecom operates its AI models using a combination of graphics processing units (GPUs) and Rebellions' Atom Max, and first deployed Atom Max in December 2025. Rebellions' AI infrastructure handles 14 million requests a day, equivalent to 4 billion tokens, a scale the company said is comparable to AI operations at global companies. Atom Max is Rebellions' NPU for data centers; NPUs are AI chips specialized for AI inference workloads, offering higher power efficiency and lower costs than GPUs, which are used for both AI training and inference. Park Byung-kwan, head of Core Platform at SK Telecom, said before government officials in April 2026 that the company "plans to continue expanding commercial services based on Atom Max." Rebellions said its NPUs provide low-latency, high-performance computing and can operate reliably for real-time services, support global-standard AI development and deployment environments, and that it optimized computing resources to match SK Telecom's AI models and improved infrastructure operating efficiency by building farms consisting of hundreds of NPUs and cards. CEO Park Sung-hyun said services used by people every day are running reliably on