AI All Funding Startup News in Korea (Korean Startup News)

SqueezeBits Raises KRW 2.5 Billion in Pre-Series A for AI Lightweight Tech

2024-01-23 2 min read


SqueezeBits Raises KRW 2.5 Billion in Pre-Series A for AI Lightweight Tech

Reading Time: 2 minutes

The South Korean AI lightweight technology startup, SqueezeBits, has successfully raised KRW 2.5 billion in a pre-series A funding round. This round included investments from Kakao Ventures, Samsung Next, POSCO CAPITAL, and POSTECH Holdings.

SqueezeBits is at the forefront of developing AI lightweight technology, designed to optimize AI service usage. This technology significantly reduces the memory and computational requirements during AI model inference, enabling swifter processing. A key aspect of their innovation is the ability to quantize 32-bit data to 4 bits or less while maintaining AI model performance, essential for creating faster and lighter AI models.

Additionally, SqueezeBits owns a software engine capable of efficiently processing these quantized models on existing hardware. In an era marked by the proliferation of AI-based services such as ChatGPT, this technology is anticipated to substantially cut down operational costs, addressing a major hurdle in AI democratization.

The versatility of SqueezeBits’ AI lightweight technology is evident in its wide range of applications, from mobile smartphones and laptops to edge devices and GPU clouds. It supports diverse models, including those for images, videos, voice, and natural language, broadening its potential use cases. The company has completed technology validation (PoC) and projects with over 20 firms, including NAVER and SKT, showcasing its lightweight technology in various AI services, especially in large-scale AI models.

SqueezeBits has also attracted attention for its on-device AI demos, including real-time operation of large language models (LLM) on edge devices and running the AI image generation model, Stable Diffusion, on smartphones.

Following this investment, SqueezeBits aims to solidify its position in lightweight technology and aggressively pursue international market expansion. The company has also launched the OwLite toolkit, allowing even those unfamiliar with lightweight technology to easily apply and analyze AI models. Currently compatible with NVIDIA GPUs, the company plans to extend its support to other hardware platforms like Intel, AMD, ARM, and NPUs from various AI semiconductor startups.

Originating from a POSTECH graduate school research team specializing in deep learning accelerator hardware (Neural Processing Unit, NPU), SqueezeBits has a solid foundation. Over the past seven years, the SqueezeBits team has consistently published papers in leading machine learning conferences such as CVPR, NeurIPS, and ICLR, totaling over 70 international papers focused on deep learning acceleration. Their experience in designing AI-specific hardware underlines their comprehensive expertise in optimizing AI algorithms and hardware.

Jung-Ho Shin of Kakao Ventures highlighted the investment’s reasoning, stating, “SqueezeBits is a team capable of universalizing AI applications, offering comprehensive solutions across the AI value chain, thanks to their expertise in both hardware and software.”

Hyung-Jun Kim, CEO of SqueezeBits, emphasized the company’s mission, saying, “To transition AI-based services from customer acquisition to revenue generation, it is crucial to drastically reduce AI operational costs. SqueezeBits is committed to resolving these cost and efficiency challenges for numerous companies, leveraging our AI lightweight technology to fully realize the potential of AI.”

[Korean AI Startups News]

스타트업 투자의 숨은 말 '다코스'
Leave a comment

댓글 남기기