HyperExcel, a domestic artificial intelligence (AI) semiconductor startup specialized in large language models (LLM), announced on the 26th that it has secured 55 billion won in Series A investment. HyperExcel is a system semiconductor startup that develops LLM-specific server solutions based on its self-developed LPU (LLM Processing Unit).
The Series A round was led by Korea Investment Partners, with participation from Korea Development Bank, KB Investment, Mirae Asset Venture Investment, SBVA, Bon Angels, Korea Development Bank Capital, LB Investment, Company K Partners, and global investor Vickers Venture Partners, in addition to domestic investors.
Kim Hee-jin, the senior team leader of Korea Investment Partners, noted, “HyperExcel is expected to provide highly efficient chips that are both cost-effective and performance-effective compared to NVIDIA by developing AI semiconductors optimized for LLMs. We are currently conducting joint development with several large corporations both domestically and internationally and expect to produce products by the end of next year.”
Through this investment, HyperExcel plans to mass-produce the world's first low-cost LLM-specific chip equipped with LPDDR5X memory, targeting a launch next year, using Samsung Electronics' 4nm (nanometer, one-billionth of a meter) foundry process. HyperExcel revealed that its LPU demonstrated about 10 times better price efficiency compared to high-performance graphics processing units (GPUs) during production as application-specific integrated circuits (ASICs).
Kim Joo-young, the CEO of HyperExcel, stated, “High-performance GPUs based on HBM have reached their limits in terms of price and power. By developing its own LPU semiconductor based on low-power memory LPDDR for the server market, HyperExcel will significantly reduce the expenses and power required for LLM inference and achieve data center efficiency.”