[AI Paradox] Investment in 'Trillion-Won Units' is Basic... 'Separating Wheat from Chaff' Amid Bubble Concerns Coming

Nakao Makes Record Investment in AI... Raises Funds by Selling Real Estate
ChatGPT Still Lacks Profit Model... AI Bubble Speculation Grows

Is it a goose that lays golden eggs or a money-eating hippo? Companies ranging from domestic conglomerates to startups that have ventured into generative artificial intelligence (AI) are bearing enormous cost burdens. Large language models (LLMs), which form the basis of generative AI, require trillions of won in funding from development to operation. Many have fallen into deficits until their revenue models are established. Nevertheless, as the 'AI rush' continues, even bubble theories are being raised. There are talks that the AI bubble is growing in size.

Trillions Invested in Development and Operation... Struggling with Revenue Models

Naver and Kakao spent 964.9 billion won and 544.7 billion won respectively on research and development in the first half of this year. With a significant increase in AI research and development, these amounts represent a 15% and 6% increase compared to the same period last year. This is the largest scale ever recorded for the first half of the year.

AI investment alone amounts to hundreds of billions of won. Developing LLMs requires computing resources such as graphics processing units (GPUs) and electricity, specialized personnel, and vast amounts of training data. Naver, which released Korea's first domestically developed LLM, HyperCLOVA, and the next-generation LLM HyperCLOVA X, has poured about 1 trillion won into AI over the past 3 to 4 years. Due to the enormous capital required, it has even resorted to selling real estate. Naver recently sold a 45% stake in the Pangyo Techno Tower to the Singapore Investment Corporation (GIC), reportedly securing 350 billion won.

Kakao is also struggling under the burden of AI development costs. Despite surpassing 2 trillion won in sales for the first time in Q2 this year, its operating profit fell by 34%. The company expects infrastructure-related costs to increase by about 50% compared to last year as AI investments continue in the second half. Although it earns a lot, the scale of investment is even larger, resulting in less remaining profit.

Even after development is completed and services are launched, expenses continue. Running AI requires massive computing resources. For example, ChatGPT, used by 1.5 billion people, reportedly incurs about 900 million won per day in cloud service fees. Since it is a new service, predicting operating costs is difficult. This is why Naver and Kakao are concerned about their revenue models. Kim Nam-sun, Naver's Chief Financial Officer (CFO), said, "It is difficult to predict the scale of operating costs for ultra-large AI services," adding, "We will closely monitor user behavior and scale in the future to implement appropriate business strategies."

Many Deficits Despite Developing LLMs... "Must Prove Marketability"

Except for large corporations, most companies endure deficits to invest in AI. AI software (SW) listed company Conan Technology, which recently unveiled its own LLM, recorded sales of 4.5 billion won and an operating loss of 7.4 billion won in the first half of this year. Vibe Company, which also owns its own LLM, posted sales of 11.5 billion won and an operating loss of 6.6 billion won. Since 2019, it has recorded deficits that continue to grow in scale.

Given this situation, AI bubble theories have emerged. Some companies jump into LLM development without clear business models, while investment money is pouring into such companies. Kwon Oh-hyung, CEO of venture capital (VC) FuturePlay, said, "There is definitely a bubble as valuations rise and investors flock thanks to AI," adding, "Since even ChatGPT, which sparked the boom, is still in the process of finding a revenue model, we need to be cautious about doing business with AI models."

There are already cases overseas where the bubble has burst. Prisma Labs in Silicon Valley, USA, launched 'Lensa AI' last year. It is an app that creates portraits using generative AI when users upload photos. Although it achieved daily sales of 1 billion won shortly after launch, its popularity quickly faded. In December last year, it ranked first in downloads on the Apple App Store and Google Play Store but is now out of the rankings. Andrew Ng, a professor at Stanford University, said about Lensa AI, "Only sustainable AI services that help humanity will survive."

There is a growing voice that generative AI must prove its marketability to avoid ending as just a bubble. Ham Myung-jin, CEO of Finders AI, said, "We considered natural language models but judged that it is a field where large corporations can excel," adding, "It must become a sharp service that clearly solves needs in a specific area." Shin Dong-ho, CEO of Muhayu, also emphasized, "If you rank third or fourth in the AI field, you only incur costs and cannot catch up with big tech companies," adding, "It is important to aim to be number one with a service and become a company that makes money." Frankly speaking, the only company currently making money from AI is Nvidia, which manufactures related hardware. AI developers themselves are like steam locomotives burning money while running toward the future.