Traders at the New York Stock Exchange [AFP=Yonhap News. Redistribution and DB prohibited]

The theme of this week's market trend can be summarized as “Earnings beat, but stocks lost.”

Looking solely at the fundamentals of the semiconductor industry, it is difficult to explain this week's sharp stock price decline.

TSMC, the world’s largest foundry, reported $40.2 billion in revenue for the second quarter, hitting the top end of its guidance of $39 billion to $40.2 billion. Its gross margin of 67.7% and operating margin of 60.3% also surpassed the upper limits of its previous projections.

TSMC’s revenue guidance for the third quarter is between $44.6 billion and $45.8 billion. Based on the midpoint, this represents an increase of more than 12% from the second quarter. Both the actual results and the outlook demonstrate that demand for high-end AI semiconductors remains robust.

However, stock prices moved in the opposite direction. The Philadelphia Semiconductor Index fell by about 10% over the week, marking its largest weekly decline in over a year. Having dropped more than 20% from its all-time high in late June, the index has entered technical bear market territory. Nevertheless, its year-to-date gains remain above 60%.

This is why this correction should not be dismissed as a “collapse in AI demand.” It is more accurate to say that future growth expectations, which had already been priced into the stocks, were simply too high rather than the current orders and sales collapsing.

AI and semiconductor stocks rose rapidly on the premise of continuous expansion in capital expenditure (CAPEX), supply shortages of high-end chips, and high profit margins. Investors have now begun to recalculate the speed at which AI infrastructure investments translate into actual revenue and cash flow.

The proliferation of low-cost, open-source models by Chinese AI firms, the payback period for data center investments, and the depreciation burden resulting from increased CAPEX have emerged as new variables. The question has grown as to whether semiconductor companies can continue to exceed the expectations already baked into their stock prices simply by meeting market estimates.

Reuters analyzed this semiconductor correction as a confluence of profit-taking, a reassessment of the sustainability of AI capital investment, and the liquidation of overvalued momentum stocks and leveraged positions.

The correction was more pronounced in the Korean market. This is because Samsung Electronics and SK Hynix account for a large portion of the KOSPI, and credit loans, futures, single-stock leverage products, and the prices of SK Hynix’s American Depositary Receipts (ADRs) are all interlinked in one direction.

On the 13th, SK Hynix and Samsung Electronics plummeted by roughly 15% and 10%, respectively. While they rebounded by 6% and 8% on the 15th, they fell again on the 16th by 11.53% for SK Hynix and 8.77% for Samsung Electronics. It is more reasonable to view this as overseas stock prices and leverage-related supply and demand amplifying domestic prices, rather than earnings outlooks changing drastically overnight.

The answer to last week's question, “Is the semiconductor correction merely profit-taking?” is that it started with profit-taking but expanded into a correction combined with the liquidation of leverage and concerns over the return on AI investments.

That said, this does not confirm the end of the memory super-cycle. TSMC’s performance is evidence that demand for advanced manufacturing processes and AI infrastructure remains alive. However, the market is unlikely to assign the same valuation to all semiconductor companies and products based solely on the explanation that “AI investment is increasing.”

The distinction between products with clear long-term contracts and supply constraints, such as High Bandwidth Memory (HBM) and server DRAM, and those sensitive to price and supply expansion, such as general-purpose DRAM and NAND, will become more critical. It has become increasingly likely that stock prices will be determined not just by revenue growth, but by return on investment, cash flow, actual data center utilization rates of clients, and product-specific profitability.

The conclusion of this week’s Stock Radar is as follows.

The stock price decline was not caused by a downturn in the semiconductor industry. Rather, stock prices and expectations had moved too far ahead of the industry's actual condition, and the liquidation of leverage abruptly closed that gap.

There are three key points to watch next week.

First, we must confirm whether the planned AI capital expenditures remain intact in the earnings reports of major U.S. tech companies.

Second, we must observe whether SK Hynix’s earnings prove the profitability gap between HBM and general-purpose memory.

Third, we must monitor whether the Philadelphia Semiconductor Index finds support at the 20% decline line from its June peak.