KOSPI plunged 20.7% in 7 trading days... Despite profit increase, corporate value is reassessed, 47 US data centers are disrupted... Gap between AI investment profitability and semiconductor performance
As artificial intelligence (AI)-related stocks and semiconductor stocks plummet, comparisons between the current stock market and the time of the dot-com bubble burst in 2000 are spreading. Even as the profit outlook for major semiconductor companies such as Samsung Electronics and SK Hynix is improving, their stock prices are falling sharply, and concerns are growing in the United States about the financing and profitability of large-scale data center investments. It is similar to 26 years ago in that the gap between corporate performance and stock prices is widening, but it is different from then in that the AI industry is already generating significant sales and profits.
KOSPI fell for 7 consecutive trading days from the 6th to the 15th, plummeting 20.7%, then rebounded 2.05% on the 16th and finished trading at 5668.05. The stock market, which has fallen by more than 20% in a short period of time, is being compared to the bear market during the past financial crisis and the collapse of the dot-com bubble. A question of interest in the market has emerged as to whether the recent adjustment will only result in a temporary decline in investment sentiment or lead to a reevaluation of the profitability of AI investments.
#### Even though profits increase, stock prices plummet... Memories of the dot-com bubble 26 years ago
According to the 'Shinhan M.R.I.' report published on the 16th by Jinhyeok Kang, a researcher at Shinhan Investment & Securities, since 2000, there have been five cases, including this time, when the KOSPI fell for more than six consecutive trading days and the cumulative decline exceeded 15%. During the 2000 dot-com bubble, it fell 17.5% over 9 trading days, and during the 2008 financial crisis, it fell 17.7% over 8 trading days. During the U.S. credit rating downgrade and European financial crisis in 2011, it fell 17.1% over 6 trading days, and during the COVID-19 pandemic in 2020, it fell 25.7% over 7 trading days.
The speed of recovery after the plunge varied greatly. As of 60 trading days after the continuous decline ended, KOSPI rose 24.4% in 2008 and 49.3% in 2020, but fell an additional 4.9% in 2000. In 2011, it only increased by 6.0%. The subsequent trend changed depending on the nature of the shock that shook the market rather than how much the stock price fell.
Shinhan Investment & Securities paid attention to the discrepancy between the company's profit outlook and stock price trend during this sharp decline. While earnings per share (EPS) forecasts were revised upward, especially in the semiconductor industry, the price-to-earnings ratio (PER) fell significantly. Rather than a collapse in performance itself, the value the market places on future profits has decreased, putting pressure on stock prices.
A similar phenomenon occurred when the dot-com bubble burst in 2000. Although companies' profit estimates increased, the decline in PER outweighed them, and even after the sharp drop, stock price recovery was delayed. On the other hand, during the financial crisis in 2008 and the pandemic in 2020, PER rebounded quickly thanks to policy responses and expectations of economic recovery even as the performance outlook worsened.
Researcher Kang Jin-hyuk said in the report, "The nature of the shock and the possibility of its resolution, rather than whether there is a sharp decline itself, determine the future direction of the stock price."
There were also trends different from the past in foreign supply and demand and exchange rates. The won-dollar exchange rate rose during the sharp declines in 2000, 2008, 2011, and 2020, but this time it actually fell amid continued foreign selling. This is different from past major plunges where Foreign Investors' stock selling led to the weakening of the won.
#### 47 U.S. data centers disrupted... Growing doubts about AI investment
Doubts about AI investment are focused on whether the enormous facility investment can actually be executed and recovered as profit rather than the potential of the technology. As the burden of financing increases amid delays in securing the power and permits required for data center construction, contracts are being withdrawn and investment schedules are being changed in some projects.
According to a report by the Financial Times (FT) on the 18th, citing a tally by regulatory information company RegLayer, at least 47 large data center projects were delayed, withdrawn, or blocked in the United States this year. Problems with power grid connection, licensing, and opposition from local communities were cited as the main causes. Bernstein, a global investment bank, also estimated that among the data center construction plans announced in the United States, the amount with a high possibility of actual realization is about 35%. This is an estimate, not a completion performance, and it is difficult to believe that all announced investment plans will lead to actual supply capacity.
In the UK, Microsoft withdrew its data center lease agreement with AI infrastructure company Nscale. It is reported that the facility will only be able to supply power in the 2030s, but Microsoft did not disclose the reason for withdrawing the contract. This is an example that shows that even if AI semiconductors and servers are secured, the project schedule may be delayed if the power and facilities to operate them are not prepared in time.
Construction delays lead to financial burdens. Data center operators must invest large amounts of money in securing land, construction, and power infrastructure before operating the facility. If the business schedule is delayed, the timing of sales generation and investment recovery will be pushed back, and the higher the proportion of borrowings, the greater the burden of financial costs for a company.
According to a study by the Brookings Institution, investment in AI infrastructure in the United States from 2025 to 2032 is estimated to reach $10.3 trillion. This is equivalent to about 3.6% of the annual average U.S. gross domestic product (GDP). Investment resources are also expanding beyond Big Tech's own cash flow to include joint ventures, private loans, and special purpose corporations.
The simultaneous decline in stock and bond prices of AI cloud company Coreweave also shows that these concerns are spreading to the financial market. As the scale of AI infrastructure investment grows, operators need to secure more funds, but as uncertainty about return on investment increases, additional financing conditions may become unfavorable.
#### Semiconductors are performing well, but the return on investment has not yet been achieved... Contrasting shades of AI
Even after the dot-com bubble burst in 2000, the Internet industry did not stop growing. However, companies that failed to secure a profit model were pushed out of the market one after another, and excessive investment in communication networks and servers increased the financial burden on related companies. The Nasdaq Composite Index fell approximately 78% from its high point in March 2000 to its low point in October 2002.
Now, semiconductor companies such as NVIDIA, TSMC, Samsung Electronics, and SK Hynix are enjoying direct benefits from expanded AI investment. As demand for high-bandwidth memory (HBM) and AI accelerators increased, sales and profits also increased. Unlike some Internet companies during the dot-com bubble, the fact that they are already making profits is an important differentiator of the current AI industry.
However, the profit structures of semiconductor suppliers and data center operators cannot be viewed as the same. Semiconductor companies secure sales by selling products, but big tech and cloud operators must first make massive facility investments and then recover the investment through AI service sales. The strong performance of semiconductor companies does not guarantee the profitability of the data center business.
The future performance of Samsung Electronics and SK Hynix is also not unrelated to the investment capacity of American big tech. If AI service sales do not meet expectations or financing costs rise, investments in new data centers may be adjusted first. This is an issue that can affect the order size and supply price of HBM and server memory.
Researcher Jinhyeok Kang warned that if the burden of AI financing leads to a reduction in facility investment and lower semiconductor profit estimates, a long-term decline in corporate value like the one in 2000 could occur despite the lower PER.
To date, there is limited evidence that data center construction delays have led to a widespread decline in semiconductor orders. It is difficult to interpret the delay in data center construction due to power grid and licensing issues as a decrease in demand for AI semiconductors. While the profit outlook for semiconductor manufacturers remains intact, the investment burden is first evident in data center operators and the financing market.
What the Internet industry left behind in 2000 was not only the achievements of technological advancement. Investment performance varied greatly from company to company, and there were many cases where corporate value, which had increased ahead of growth expectations, failed to recover. In the AI industry, it is increasingly likely that how quickly massive facility investments are converted into actual cash flow and profits will determine the stock price trends of each company.
