Funding secured for AI infrastructure startup

An early-stage venture capital firm has completed a seed investment in an AI infrastructure software startup. The startup, focused on developing technology to reduce artificial intelligence inference costs, aims to lower these costs by 40 to 70%. The investment was finalized on October 22, with the specifics of the amount undisclosed, although the company is set to receive up to 800 million KRW in research and development funding through the TIPS program, as announced on October 21.

The startup, co-founded by professors from a prominent academic institution, is creating a multi-accelerator execution environment that integrates various AI semiconductors. The technology combines graphics processing units (GPUs), neural processing units (NPUs), and processing-in-memory (PIM) to optimize inference tasks. The startup is also receiving advice from a co-founder of a major AI semiconductor company who is involved in the project.

As generative AI services become more widespread, the costs and processing efficiency of operating these models have emerged as critical challenges. Current large language model (LLM) services rely heavily on GPUs, but leveraging a variety of AI semiconductors necessitates software environments tailored to the specific characteristics of each hardware type.

The startup's multi-accelerator runtime automatically allocates the most suitable semiconductors for inference tasks based on factors such as response speed and cost. The company aims to significantly reduce inference costs compared to infrastructures that solely utilize GPUs. They are also ensuring compatibility with existing development environments.

The venture capital firm expressed optimism that this technology will alleviate operational cost burdens for companies adopting AI solutions. The startup plans to conduct proof-of-concept trials with several clients by the end of the year and aims to officially launch its product in 2027 after validating its technology with a cloud service provider.

Reported by StartupN.