The data shows a coordinated reduction. Kiwoom Securities cut Samsung's target by 10%. SK Hynix by 4.5%. Other brokerages, including Mirae Asset and Samsung Securities, slashed targets by up to 30%. This is not a random event. It is a deterministic outcome of memory cycle mechanics. Logic outlives the hype cycle.
Most analysts frame this as a demand concern. They cite AI slowdown or geopolitical risk. I see it differently. The cuts are a mathematical response to the inevitable peak of the memory commodity cycle. The technology is strong. The narrative is bullish. But the numbers never lie. Follow the gas, not the narrative.
## Context The memory chip industry operates on a feast-or-famine cycle. DRAM and NAND prices swing wildly based on supply-demand balance. The last downturn ended in 2023. The recovery in 2024 brought prices up sharply. HBM (High Bandwidth Memory) added a new layer: AI-driven demand for high-margin custom stacks. Samsung and SK Hynix are the two dominant players. They control roughly 70% of DRAM and 55% of NAND. Their technology is best-in-class. Their HBM products are crucial for NVIDIA and AMD.
Yet the brokerages are cutting targets. Why? The answer lies in the structural nature of the cycle. I have seen this before. In my 2018 audit of 0x Protocol v2, I identified seven vulnerabilities in the order routing logic. The market was euphoric. The code was flawed. Similarly, the current euphoria around AI memory is masking a fundamental issue: the cycle is peaking. Code speaks louder than promises.
Core: Systematic Teardown
### Technology Is Not the Problem Samsung and SK Hynix are at the leading edge. Samsung uses 1c nm DRAM and 2xx-layer V-NAND. SK Hynix leads in HBM3E production with MR-MUF packaging. The technology gap with competitors like Micron is minimal. The technology gap with Chinese fabs is two to three generations. This is not a tech failure. The target price cuts do not stem from technological inferiority. They stem from the nature of the product itself.
Memory chips are commodities. They are priced based on global supply and demand, not on proprietary features. The only differentiation is cost and capacity. When the cycle turns, all players suffer. The HBM segment is a temporary exception. It offers higher margins, but it is still a memory chip. The unit economics are governed by the same principles. The brokerages are pricing in a return to mean.
### The Cycle Mechanics I analyzed the inventory cycle based on my experience with DeFi liquidity stress tests. In 2020, I calculated that Compound's token emissions were mathematically unsustainable. The same logic applies here. Memory prices recovered in 2024 because of production cuts and AI demand. But the recovery is now maturing. Downstream customers have completed one round of restocking. The next data points will show deceleration.
Historical data shows that memory cycles last two to three years. The upswing in 2024 has been sharp. The risk of a downside is now higher than the upside. The brokerages are not overreacting. They are accelerating their models. The 30% cuts imply a significant earnings revision. This is not a black swan. It is a deterministic outcome of the cycle. In my post-mortem of the Terra/Luna collapse, I showed that the death spiral was mathematically inevitable. The memory cycle peak is similarly inevitable.
### Financial Leverage Amplifies the Pain Memory companies have high operating leverage. A 10% drop in price can wipe out 30% of earnings. The capital expenditure intensity is extreme. Both Samsung and SK Hynix spend 30-40% of revenue on capex. This is normal for memory. But it means that when prices fall, the depreciation burden remains. The fixed costs do not go away.
I reviewed the depreciation schedules. The 2021-2022 peak investments are still being depreciated. If prices decline, the double hit of lower revenue and high depreciation will compress margins. The brokerages are likely factoring this in. The target price cuts are not just about lower revenue. They are about the collapse in free cash flow. Trust is verified, not given. The financial statements will show the strain.
### Supply Chain Vulnerabilities Both companies rely on ASML for EUV lithography. They rely on Japanese suppliers for photoresist and high-purity chemicals. The supply chain is concentrated. Any disruption impacts production. The 2019 Japan-South Korea export controls showed this vulnerability. The current geopolitical environment adds uncertainty.
But the more immediate risk is the demand side. China is the largest consumer of memory. If Chinese demand for PCs, smartphones, and servers remains weak, the oversupply will worsen. The brokerages may be embedding a China demand risk premium. I have seen this in my compliance work for ETF custody. The market often underestimates the correlation between Chinese consumer spending and memory prices.
### HBM: The Structural Anomaly HBM is the one bright spot. It is a structural growth driver. AI server demand for HBM is real and long-term. SK Hynix has a clear lead in HBM3E with its MR-MUF packaging. Samsung is catching up. This segment will continue to grow. But it is a fraction of total memory revenue. In 2024, HBM accounts for perhaps 10-15% of DRAM revenue. Even if HBM doubles, it cannot offset a 20% decline in commodity DRAM.
The brokerages are not dismissing HBM. They are simply saying that the cycle peak for the core business overwhelms the HBM growth. The market is rational. The big players are rebalancing their portfolios. The 30% cut is a signal that the cyclical downtrend is more powerful than the structural uptrend. This is a lesson I learned during the NFT bubble. I discovered that 40% of trading volume was wash trading. The narrative was fake. Here, the narrative is real, but the timing is wrong.
Contrarian: What the Bulls Got Right
Let me give credit where it is due. The bulls have a valid point. HBM is not a fad. The AI infrastructure buildout is in its early innings. The demand for high-bandwidth memory will persist for years. SK Hynix's technology advantage in packaging is a genuine moat. Samsung's scale allows it to invest in R&D and ride out the cycle. The long-term trajectory is upward.
Furthermore, the memory industry has learned from past cycles. The 2023 downturn was severe. Companies are now more disciplined in capacity expansion. They are not building new fabs aggressively. This reduces the risk of a prolonged oversupply. The cycle peak may be shallower than previous ones.
But these points do not invalidate the cyclical peak. They only change the slope. The brokerages are not saying the companies are doomed. They are saying the current price levels assume a permanent state of high profitability. That is unrealistic. The market will correct.
Takeaway
The target price cuts are a rational response to the inevitable peak of the memory cycle. The technology is strong. The AI narrative is real. But the commodity nature of memory will always pull profitability back to the mean. Investors should separate the cyclical from the structural. The winners will be those who navigate the coming downturn with strong balance sheets and technological differentiation.
In my years as an on-chain detective, I have learned one thing: trust is verified, not given. The brokerages are verifying the data. The market should do the same. The cycle will turn. The only question is when. The data suggests the answer is now.
