KOSPI plunges 20.7% in seven sessions... Corporate valuations reassessed despite rising earnings
47 U.S. data center projects hit snags... The widening gap between AI investment returns and semiconductor performance
As artificial intelligence (AI)-related and semiconductor stocks suffer a sharp downturn, comparisons between the current market and the 2000 dot-com bubble are gaining traction. Even as profit forecasts for major semiconductor companies—including Samsung Electronics and SK hynix—improve, their share prices have plunged steeply. Simultaneously, growing concerns are emerging in the U.S. over the financing and profitability of massive data center investments. While this disconnect between corporate earnings and stock prices mirrors the situation 26 years ago, today's AI industry stands apart in that it is already generating substantial sales and profits.
After dropping 20.7% over seven consecutive trading sessions from January 6 to 15, the KOSPI rebounded 2.05% on the 16th to close at 5,668.05. A more than 20% decline over such a short span has drawn parallels to past bear markets during financial crises and the dot-com crash. Market attention has now turned to whether this recent correction will remain a temporary cooling of sentiment or lead to a fundamental re-evaluation of AI investment returns.
Surging Profits, Plunging Stocks: Memories of the Dot-Com Bubble 26 Years Ago
According to a report published on the 16th by Kang Jin-hyuk, an analyst at Shinhan Securities, there have been five instances since 2000—including the current episode—where the KOSPI fell for six or more consecutive trading sessions with a cumulative drop exceeding 15%. During the dot-com bubble in 2000, the index dropped 17.5% over nine sessions; during the 2008 financial crisis, it fell 17.7% over eight sessions. It declined 17.1% over six sessions during the U.S. credit rating downgrade and European debt crisis in 2011, and dropped 25.7% over seven sessions during the COVID-19 pandemic in 2020.
The speed of recovery following these steep declines varied dramatically. Measured 60 trading days after the consecutive drops ended, the KOSPI rose 24.4% in 2008 and 49.3% in 2020, but fell an additional 4.9% in 2000. In 2011, it managed a meager 6.0% gain. The subsequent trajectory depended less on the magnitude of the sell-off and more on the nature of the shock that shook the market.
Shinhan Securities highlighted the divergence between corporate earnings forecasts and stock price trends during the current downturn. While consensus estimates for earnings per share (EPS), led by the semiconductor sector, have been revised upward, price-to-earnings (P/E) ratios have fallen sharply. Rather than a collapse in fundamentals, it is a contraction in the valuation the market assigns to future profits that has weighed on share prices.
A similar phenomenon occurred during the 2000 dot-com crash. Corporate earnings estimates rose, but the contraction in P/E multiples far outweighed them, and stock price recovery lagged even after the crash. Conversely, during the 2008 financial crisis and the 2020 pandemic, P/E ratios rebounded rapidly on the back of policy responses and economic recovery expectations, despite deteriorating earnings outlooks.
Kang noted in the report, "The nature of the shock and its potential for resolution—rather than the plunge itself—determines the subsequent direction of stock prices."
Foreign capital flows and exchange rates have also exhibited a pattern distinct from the past. During the crash periods of 2000, 2008, 2011, and 2020, the won-dollar exchange rate rose; this time, however, the exchange rate has fallen despite persistent foreign selling. This marks a departure from past major downturns, where foreign equity liquidations drove weakness in the Korean won.
47 U.S. Data Center Projects Face Hurdles: Mounting Doubts Over AI Investment
Skepticism over AI investment centers less on the technology's potential and more on whether massive capital expenditures can actually be executed and recouped through profits. As securing power and permits for data center construction is delayed—compounded by escalating financing burdens—some projects are seeing contract cancellations and investment timeline adjustments.
According to Financial Times reports citing data from regulatory intelligence firm RegLayer, at least 47 major U.S. data center projects have been delayed, canceled, or blocked this year. Grid connection issues, permitting hurdles, and local community opposition were cited as the primary causes. Global investment bank Bernstein also estimates that only about 35% of announced U.S. data center construction plans have a high probability of materializing. This figure represents a forward-looking assessment rather than completed projects, suggesting that announced investment plans will not easily translate into actual supply capacity.
In the UK, Microsoft canceled a data center lease agreement with AI infrastructure firm Nscale. While power supply to the facility was reportedly not expected until the 2030s, Microsoft did not disclose the reason for the cancellation. This case illustrates that even if AI chips and servers are secured, project timelines can slip if the power and facilities required to run them are not ready on schedule.
Construction delays translate directly into financial strain. Data center operators must inject massive capital into site acquisition, construction, and power infrastructure buildup well before facilities become operational. Delays push back the generation of revenue and the recovery of investment capital, increasing financial costs—particularly for highly leveraged companies.
Research by the Brookings Institution estimates that U.S. AI infrastructure investment between 2025 and 2032 will reach $10.3 trillion, equivalent to about 3.6% of average annual U.S. GDP. The financing sources for these investments are expanding beyond Big Tech’s internal cash flows to include joint ventures, private credit, and special purpose vehicles.
The concurrent decline in the share and bond prices of AI cloud provider CoreWeave shows that these concerns are spilling over into financial markets. As the scale of AI infrastructure investment grows, operators must secure increasingly large amounts of capital, but rising uncertainty over investment payback could lead to harsher financing conditions.
Strong Semiconductor Earnings, But Payback Still Pending: The Dual Face of AI
Even after the collapse of the 2000 dot-com bubble, the internet industry did not stop growing. However, companies unable to secure viable business models were successively squeezed out of the market, and overinvestment in telecommunication networks and servers amplified the financial burdens of related firms. The Nasdaq Composite Index plummeted roughly 78% from its peak in March 2000 to its trough in October 2002.
Today, semiconductor heavyweights such as Nvidia, TSMC, Samsung Electronics, and SK hynix are reaping the direct benefits of expanding AI investments. Surging demand for high-bandwidth memory (HBM) and AI accelerators has driven up both sales and profits. Unlike certain internet companies during the dot-com bubble, today’s semiconductor firms are already generating profits, marking a crucial distinction for the current AI industry.
However, the revenue structures of semiconductor suppliers and data center operators cannot be viewed identically. Semiconductor firms secure revenue by selling products, whereas Big Tech and cloud providers must first execute massive capital expenditures and then recoup those outlays through AI service revenues. Robust earnings by chipmakers do not guarantee the profitability of data center operations.
The future earnings of Samsung Electronics and SK hynix are also inextricably linked to the investment capacity of U.S. Big Tech. If AI service revenues fall short of expectations or financing costs rise, investments in new data centers could be the first items on the chopping block. This could ultimately impact order volumes and supply prices for HBM and server memory.
Kang warned that if AI financing burdens translate into reduced capital expenditures and downward revisions to semiconductor profit estimates, a prolonged decline in corporate valuations similar to 2000 could materialize, despite lower P/E multiples.
To date, evidence that data center construction delays have translated into a broad-based decline in semiconductor orders remains limited. It is difficult to equate construction delays caused by power grid and permitting bottlenecks with a drop in demand for AI chips. While earnings forecasts for semiconductor manufacturers remain intact, investment pressures are first manifesting among data center operators and in the financing markets.
The internet industry of 2000 left behind more than just technological advancements; investment outcomes varied wildly among companies, and many failed to recover valuations that had run ahead of growth expectations. In the AI industry as well, how rapidly massive capital expenditures translate into actual cash flow and profits will ultimately determine corporate winners and losers.
