After Samsung Electronics announced its second-quarter earnings guidance on July 7, the stock price plunged by nearly 10% intraday, dragging down the KOSPI index and triggering trading halts or circuit breakers.
The guidance itself was not bad. Samsung's announcement revealed that the second-quarter consolidated sales were approximately 171 trillion Korean won, with an operating profit of around 89.4 trillion Korean won. Compared to the operating profit of 4.68 trillion Korean won from the same period last year, this represented an almost 19-fold increase. According to LSEG SmartEstimate cited by Reuters, the market's previous expectation was around 87.3 trillion Korean won, below Samsung's guidance.
The market's reaction indicates that investors are no longer questioning whether there is AI demand but rather how long this round of high profitability can last. Samsung and SK Hynix have already felt the impact of AI server demand reassessment, prompting the market to recalibrate the profit trajectory for the second half of 2026 and 2027.
Earnings Not a Bust; AI Memory Demand Still Materializing
Samsung's guidance was not a disappointment. Instead, it was more of a reflection of AI server demand continuing to overflow onto the memory factory's profit sheet.
AI servers require a large amount of high-speed memory, with HBM (high-bandwidth memory for AI chips) being the most prominent. It serves as a high-speed data channel around the GPU, determining the efficiency of chip data retrieval and processing of massive datasets. Over the past two years, the tight supply of HBM and server DRAM has driven up memory prices and profit margins.
Samsung's second-quarter operating profit soaring nearly 19-fold from the same period last year can be attributed to two factors. One is the significantly low base effect from the same period last year, and the other is the continuous increase in AI server demand post-2025. The combination of these two factors has made the year-on-year figures appear extreme.
From a current operational perspective, the prosperity of AI memory has not yet ended. Memory prices, especially those related to AI such as DRAM and HBM, are seen as the primary support for profit improvement. The complete financial report has not disclosed the segment structure, so the guidance can only confirm the total amount and cannot directly pinpoint the contribution of each product line.
Cyclical stocks often experience this kind of reaction. The profit announcement may be excellent, but the stock price still drops. Investors are buying into future profit changes, not the numbers already reflected in the guidance.
The Market Begins to Scrutinize the 2027 Profit Trajectory
Samsung was sold off, not because the market suddenly rejected AI, but because it started to doubt whether AI infrastructure spending could continue to accelerate.
The previous semiconductor upswing had an implicit assumption: large tech companies would continue to increase AI data center capital expenditure, with GPUs, HBM, and server memory all in a long-term scarcity. As long as this assumption holds, the high prices and profits of memory factories can continue to be extrapolated.
Now the question has been pushed further out. What if cloud providers' capital expenditure growth slows in the second half of 2026? What if more training needs shift to inference, leading to a decrease in per unit computing power and per unit memory requirements? What if custom chips improve efficiency, weakening the demand intensity for standard GPUs with HBM?
Clues about Zhipu AI become sensitive here. According to The Information, Zhipu or Z.ai is in preliminary contact with domestic chip design companies due to a surge in GLM-5.2 usage, exploring custom ASICs (dedicated AI chips). Some of its models are also reported to be adapted or using domestically produced computing power such as Huawei Ascend.
These pieces of information do not prove that GPU or HBM demand has been replaced. It is more like a long-term variable: AI companies will seek more customized computing solutions for cost, supply security, and inference efficiency.
Such changes will not overturn Samsung's second-quarter profit in the short term but will affect how much valuation the market is willing to give. As long as investors start to believe that the future decline in memory demand per dollar of AI expenditure is possible, memory stocks will shift from being priced on a supply-demand mismatch to being scrutinized at a cyclical peak.
Leveraged Products Amplified the Decline
If it were just performance realization, Samsung might have only experienced a normal pullback. What made the market reaction more intense is the leverage positions accumulated in the Korean market over the past few months.
Since May 2026, Korea has introduced 2x leveraged products linked to Samsung and SK Hynix. These ETFs amplify the intraday fluctuations of a single stock. For example, a 2x product linked to Samsung theoretically could drop around 10% if Samsung falls by 5% in a day.
More troubling are the daily reset and rebalancing mechanisms. When the market continuously fluctuates in the opposite direction, the net asset value erosion of the product will widen, and the funds may passively adjust positions. Korean financial regulators Lee Chan-jin and the Bank of Korea have both previously warned that these products could amplify volatility and create herd effects.
The South Korean market also has an amplifier, with Samsung and SK Hynix carrying significant weight in the index. A drop in the semiconductor leader would drag down the KOSPI. This index decline would then trigger further risk control and passive selling, with leveraged ETFs adjusting their positions in response to net asset value changes, creating a negative feedback loop.
Therefore, a sharp decline in the KOSPI and the trading halt mechanism should not be simply interpreted as Korean investors lacking faith in Samsung's performance. A more accurate explanation is the profit-taking under high expectations, ongoing concerns about AI spending sustainability, and concentration of leveraged funds, all contributing to increased volatility.
This also explains why SK Hynix is under pressure as well. Even though it holds a stronger position in HBM, the market sell-off is not just targeting a single company but rather the most crowded end of AI memory trading.
Demand Validation Ends, Sustainability Validation Begins
This event can most easily be misinterpreted as the end of the AI cycle. However, existing evidence does not yet support such a conclusion.
A more accurate statement would be that the AI memory cycle is still strong, but the market is beginning to demand more evidence to prove its continuity. Samsung's second-quarter profit announcement confirms the actual demand, and the price environment still provides support. It's just that when expectations are already high, investors will pay more attention to marginal changes.
The capital expenditure execution of large tech companies will directly affect this trade. If cloud vendors continue to increase their AI data center budgets in the second half of this year and in 2027, the memory shortage could persist, supporting the profit outlook for Samsung and SK Hynix. If the budget slows down, memory stocks may resemble more typical cyclical assets, reaching their peak earlier.
The profit quality in Samsung's full financial report will also become more important. The market needs to see the shipment proportion of HBM, server DRAM, price sustainability, and whether there are any one-time factors affecting profit margins. Guidance provides the total picture, but the complete financial report will give the structure.
Custom chips are another focal point. Explorations like Zhipu currently remain a forward-looking variable, unable to prove a turnaround in memory orders. However, if more and more AI companies reduce unit computing costs with self-developed or localized chips, the market will continue to revise downward its long-term extrapolation of standard GPUs and high-end memory combinations.
Samsung is being sold not because of poor performance, but because good performance is no longer sufficient to address new questions. The AI memory trade is transitioning from demand validation to sustainability validation, and the next stock price recovery will require not just high year-on-year profit growth, but also evidence of orders, prices, and capital expenditure continuing to support this trajectory.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
