Hyper Accel, a Korean startup specializing in AI semiconductors for Large Language Models (LLM), has successfully raised KRW 55 billion (approximately USD 41.2 million) in a Series A funding round. Founded in January 2023, the company is pioneering the development of server solutions specifically for LLMs, utilizing its proprietary LLM Processing Unit (LPU).

This funding round was led by Korea Investment Partners and featured a mix of significant Korean investors, including Korea Development Bank, KB Investment, Mirae Asset Venture Investment, SBVA, KDB Capital, BonAngels Venture Partners, Company K Partners, LB Investment alongside international participant Vickers Venture Partners.

Investors have commended Hyper Accel for its rapid progress, notably with the launch of its LLM-dedicated server, Orion, and the commercialization of its LPU-IP last November. They recognize the company’s potential to influence the growing market for AI inference servers focused on LLMs.

Kim Hee-jin, Senior Team Manager at Korea Investment Partners, remarked on the investment, stating, “We believe Hyper Accel will deliver chips that are more cost-effective and energy-efficient than those of competitors like Nvidia, thanks to their specialized LLM-optimized AI semiconductors. We are also enthusiastic about their planned product chip production by late 2025, supported by ongoing collaborations with numerous national and international enterprises.”

With the latest capital infusion, Hyper Accel is gearing up to mass-produce the world’s first low-cost LLM-specific chips that integrate LPDDR5X memory, employing Samsung’s 4-nanometer process next year. Hyper Accel’s LPU is anticipated to offer ten times the cost-efficiency in LLM inference performance compared to high-performance GPUs, potentially altering the landscape of the AI server market dominated by Nvidia.

Professor Kim Joo-young of KAIST, who founded Hyper Accel, based on his successful experience with server hardware accelerators at Microsoft in the USA, leads the company. He plans to leverage his role as director of the KAIST AI Semiconductor Systems Research Center to propel Hyper Accel to the forefront of the global LLM-specific AI semiconductor market. 

Professor Kim emphasized the advantages of their approach: “By developing our own LPU semiconductors based on low-power LPDDR memory, Hyper Accel aims to dramatically reduce the costs and energy consumption of LLM inference, optimizing data center efficiency and establishing a leadership position in the specialized semiconductor market.”