[AI Revolution](74) The Day After Tomorrow, Nvidia's Dominance Will Change the AI Market Paradigm

NVIDIA 'CUDA'-Surpassing Optimization Solution
USD 22 Million Series B Investment from US AMD and Others
"Future AI Technology Ally... Full-Scale Overseas Expansion Next Year"

A domestic startup has thrown down the gauntlet to Nvidia, which holds 95% of the AI semiconductor market. AI infrastructure solution company 'MOREH' is changing the market paradigm with AI computation automation and optimization software.

Currently, in the AI computing infrastructure market, Nvidia's GPUs dominate 95% of the computation semiconductor sector. Despite GPU prices soaring sharply, there are no competitors, so prices are set at will. Yoon Do-yeon, CEO of MOREH, said, "It's not that Nvidia's GPU performance is exceptionally superior. AMD, Graphcore, Rebellion, and FuriosaAI chips are also excellent." He believes the core lies not in hardware but in software. He explained that software like 'CUDA,' which enables efficient operation of deep learning models such as ChatGPT on GPUs, is why GPUs dominate the market. CEO Yoon, formerly of consulting firm Bain & Company, co-founded MOREH in 2020 with Dr. Jo Kang-won, a computer engineering PhD from Seoul National University.

GPUs were originally developed as graphics processing units. However, when used for weather forecasting and other applications, they demonstrated outstanding simultaneous and parallel data processing capabilities. CUDA, software created by Nvidia, was developed to efficiently utilize GPUs in these parallel processing algorithms. During the deep learning boom in 2015, many companies achieved good results using CUDA, enabling Nvidia to gain market leadership. Just as no matter how convenient a messenger app is in Korea, it cannot overcome the entry barrier of the KakaoTalk ecosystem, CUDA has become the KakaoTalk of the AI infrastructure world.

However, the emergence of large language models (LLMs) like ChatGPT has changed the situation. CUDA requires manual distribution of GPU computations, but LLMs need more computing power and more complex calculations than before, making manual optimization difficult. Simply put, whereas previously one GPU was sufficient, now 1,000 GPUs are needed. GPU prices have surged dramatically, and even if the cost can be covered, the amount of manual work has increased enormously. Especially when the data volume changes, a new computation process must be applied, which alone takes 2 to 3 months.

Yoon Doyeon, More CEO
Photo by More

MOREH has focused on this change. Since infrastructure costs have skyrocketed by using only GPUs and CUDA, creating a solution like CUDA that can be used on other companies' AI semiconductors can reduce costs. Also, automating the computation process work, which was previously done manually, with parallelized optimization software reduces time and costs. When using CUDA, if the data volume changes, a 2 to 3-month manual optimization period is required to apply a new computation process, but MOREH has automated this task through software. It can be completed in as little as 2 to 3 days. CEO Yoon said, "We are currently discussing applications with US semiconductor company AMD and other overseas system semiconductor companies," adding, "We can reduce costs by 50 to 70% compared to existing methods and cut development time in half." He further stated, "We provide infrastructure software at a level similar to or better than CUDA, offering the same user experience without being dependent on specific hardware vendors like Nvidia."

CEO Yoon defined MOREH as a facilitator of the AI industry. He said, "MOREH is like a reliable friend supporting the rapid emergence of future AI technologies like ChatGPT." He explained, "Without MOREH, only a few companies worldwide that create ultra-large AI models will survive and monopolize the market."

Although founded in 2020, the startup is expected to achieve sales exceeding 30 billion KRW this year. It has major clients such as KT and recently attracted $22 million (approximately 29.5 billion KRW) in Series B funding from AMD, KT, Forest Partners, Smilegate Investment, and others, gaining attention domestically and internationally. Based on this, the company plans to begin full-scale global expansion next year. CEO Yoon stated, "Until now, we were in the development phase refining our products; now we are entering the full business phase," adding, "Attracting overseas customers, especially from the US, is a key plan for next year."