Generative AI service products such as ChatGPT and Midjourney are attracting attention in the market. Amidst this, domestic AI experts and industry insiders gathered in one place.

On the 22nd, the Korea Artificial Intelligence Industry Association (KAIIA) held the 'Gen-AI Summit 2023' at COEX in Samseong-dong, Seoul. This two-day event is designed to foster and advance generative artificial intelligence (AI), which is expanding across the market.

On the first day, the event began with a keynote speech by Kim Jun-ha, head of the Artificial Intelligence Convergence Business Unit, followed by presentations from big tech companies that are actively developing and servicing ChatGPT, such as Microsoft, Google, and AWS, as well as from the fabless industry that develops AI semiconductors, such as △Seungjae Yoo, CEO of Persona AI, on 'Changes in NLP Brought by ChatGPT' and △Junho Baek, CEO of Furiosa AI, on 'Building High-Performant, Energy-Efficient and Scalable AI Inference Chips for ChatGPT'.

Kim Jun-ha, the CEO, predicted, “2023 will be the first year of the AI war, and amidst the giant language model competition between Google and MS, ChatGPT Plug will enter the market as a new game changer.” He also expected that there would be opportunities in AI semiconductors and accelerators that back up AI models.

On this day, Baek Jun-ho, CEO of Furiosa AI, is a representative domestic fabless company that makes AI semiconductors. Baek, who is dedicated to developing general AI semiconductors to catch up with Nvidia, which has taken over most of the global market in hyper-scale AI, was confident that “the demand for chips in the AI sector will increase explosively in the future.”
As AI services become more advanced, they inevitably face the problem of increasing costs for infrastructure construction. NVIDIA's current A100 and H100 GPU products have the high performance and computing power required for AI operations, but there are voices saying that even hyperscalers cannot help but worry about profitability compared to the cost burden as they advance to the next generation due to the increase in total cost of ownership, including high power consumption and construction costs.

Accordingly, CEO Baek predicted that, apart from the increasing number of companies introducing AI solutions, the market needs will be focused on maximizing efficiency and performance within a given system, such as 'energy saving', and predicted the direction of technological development.

He also emphasized that the point of expansion of AI semiconductors in domestic development lies in ‘usability’, adding that the key will be how to support the development environment and resolve ‘programmability’.

NVIDIA has already established CUDA and is presenting a support ecosystem, and securing a development community that supports representative deep learning frameworks such as TensorFlow and PyTorch is becoming an essential survival strategy for commercialization.

Furiosa AI is currently developing a second-generation chip, and it is reported that it will present a general-purpose AI chip solution that supports various AI models such as LLM and Vision AI based on more powerful hardware specifications than the first-generation chip.