This article may be all wrong from start to finish.

A huge amount of blind money or zero interest rates are needed.

- This article may be all wrong from start to finish.

You can often hear the abbreviation LLM around you, which you were not familiar with until last year. LLM may not be AI, but so far, LLM seems to be representing AI. LLM is said to be a technology with great potential.

However, there are people who have doubts about how to generate revenue to continue running LLM. It creates videos and images. It writes novels and poems. Enter figures into Excel and analyze them and create a chart. Based on this, it creates a PPT. Honestly, I don't have 100% of my favorite results yet. There's an exclamation of 'Wow', but this is an exclamation of how AI technology has developed like this, not an admiration of wanting to pay a lot of money for that 'result'. Of course, it will develop over time.

It is said that it can be applied to various fields, but it is also difficult to guarantee 100% accuracy. For example, Elon Musk said that autonomous driving will be done every year from 2020. I keep saying that Robo Taxi will come out again this year, but I can't believe it. It is questionable whether Robo Taxi can be implemented only by improving the component installed in the vehicle. For fully autonomous driving, I think it is necessary to innovate the transportation system across the country, not improve the performance of automobile parts.

It is said that the scope of AI application can be expanded and introduced into power generation, transmission/distribution facility operation, and power demand forecasting, but there is no choice but to carry a certain part of risk. Moreover, demonstrating the power system's 'demonstration' itself is a big adventure. How big of an adventure would it be to prove it based on the national power grid? China's global technology in HVDC and China's global technology in heavy and electric fields such as Shanghai Electric were not due to China's 'characteristics' that enabled it to withstand some of the power system risks. In other words, when the performance of AI is not 100% sure, can it be actually introduced into the power grid management system?

It could reach 99 percent by investing 100 won to develop an AI service model. It would be possible to reach 99 percent through an increase in the LLM variable, but how much more would need to be invested to achieve 100 percent? 200 won? 300 won? 1,000 won? Moreover, it is difficult to clearly understand how LLM gives such an 'answer' so it is also difficult to check whether the current performance is 99 percent or 100 percent.  

Although many LLM-based services must be developed and distributed to make profits for AI service companies, the cost of the service development process is currently burdensome. This is because LLM is getting larger and the price of semiconductors required to drive LLM is still high. Power costs cannot be ignored. AI-related semiconductors become common and energy efficiency must be greatly increased to facilitate the development and distribution of AI services.

Large-scale facility investment must be preceded in order to solve this problem. However, is there a company that is trying to bear the burden of large-scale investment to build the material foundation of AI? Will AI semiconductor-related Fabless companies or producers such as TSMC, Samsung Electronics, and Intel come forward? In order for these companies to be motivated to resolve the semiconductor shortage by investing billions and tens of billions of dollars first, I think AI service companies or AI semiconductor Fabless companies should provide a certain amount of 'money' for investment.

However, LLM-based profit models have not yet been established. AI service companies have weaker funding power. Naturally, they cannot afford to invest billions or tens of billions of dollars. As the CEO of OpenAI travels the world, he talks about the need for astronomical investment in building semiconductor production facilities, but can he actually raise that money?

It seems that the AI industry could eventually fall into the same 'Chicken & Egg Problem' as the hydrogen and electric vehicle industries. Infrastructure construction is essential to expand the demand for hydrogen and electric vehicles, but individual companies cannot take the burden of infrastructure construction alone at a time when there is a risk that demand for hydrogen and electric vehicles will not increase. In other words, the AI industry seems to be in a similar situation.

In order for various services that can generate revenue based on LLM to spread and expand demand, the cost required for service development and operation must be significantly lowered. For this, the price and power cost of AI semiconductors, which are the basis of AI service operation, must be significantly lowered. However, there seems to be no company to bear the risk for this. The government cannot even step up.

In addition, if the LLM operating cost burden is too large and it is difficult to solve the problem due to improved semiconductor performance, the S/W side should solve the problem. This can be another important factor to consider that makes it difficult for AI semiconductor-related companies to make large-scale CAPEX expansion decisions. This is because facilities may become useless due to S/W innovation (?) although large-scale CAPEX has been done.

It seems that it is difficult for many participants in the AI industry to make a decision on whether to do this or that.

The construction of a large-scale communication network that enabled the Internet revolution in the late 1990s was possible due to the 'Dot.com Bubble'. At that time, SK Telecom was the overwhelming No. 1 market capitalization in Korea. The expansion of demand for electric vehicles by Tesla in 2020 was also based on new and renewable energy generation facilities and electric vehicle charging stations that increased by several hundred GWp every year at the zero interest rate according to Pandemic.

I think the AI industry also needs this level of 'blind money' or 'zero interest rate' to build a material foundation for growth. I'm not talking about the market capitalization of a few Fabless companies increasing to a few trillion dollars and the value of unlisted Fabless companies increasing to an outrageous level. It seems that there should be a level of money that exceeds this.

However, concerns about inflation, especially in the United States, are resurfacing, and there are growing predictions that it will not be easy to cut interest rates within this year. In this situation, if huge amounts of funds are released and a bubble occurs, the inflation problem could become even worse. It seems that the time has come when more people think that AI-related companies need to actually show something.