PuriosAI, NPU Production to Double to 40,000 Units Next Year... Challenging NVIDIA's Monopoly

Korean AI semiconductor startup PuriosAI has entered mass production for large language model (LLM) inference chips, aiming to challenge NVIDIA's dominance in the AI chip market with cost-effectiveness and power efficiency. The company plans to scale production of its second-generation LLM inference chip 'RNGD' to 40,000-50,000 units by next year, doubling this year's output. This marks the first time PuriosAI's production plans have been disclosed. Rapid AI market changes, including the rise of autonomous AI and increased power consumption, are driving demand for efficient inference chips. RNGD's 200W power consumption matches high-performance GPUs, enabling twice the token generation capacity at similar power levels, significantly reducing data center TCO. These cost savings are driving demand from major clients, with RNGD adoption accelerating in South Korea. PuriosAI is also advancing long-term R&D investments, including a 2nm-based third-generation 'Stork' chip, and partnering with Broadcom to overcome NVIDIA's technical barriers through high-speed inter-chip connectivity.