The headline numbers are a lie.
Samsung Electronics is set to report a Q2 operating profit of KRW 85 trillion (USD 63 billion) on revenue of KRW 169 trillion. A 50% profit margin. In an industry where a 25% margin is considered a strong year.
Here is the disconnect: The market is pricing this as a structural renaissance. My model says this is a pulse-driven peak — a perfect storm of AI-induced HBM price hikes crashing into last year's crypto-winter lows. The real story is not the profit figure. It's what Samsung is doing with it.
They are burning cash on foundry.
The narrative is always "HBM4," "2nm GAA," "AI chip leadership." But the core tension is this: Samsung is using a single, cyclical cash cow (memory) to fund a structural money pit (foundry). Let's break the numbers down from a trader's perspective, not a sell-side analyst's.
The Context: A War on Two Fronts
The semiconductor industry operates in two distinct thermodynamic states: Speed (foundry) and Scale (memory). Foundry is about bleeding-edge process refinement: 3nm, 2nm, GAA. It requires a deep ecosystem of EDA tools, IP blocks, and customer trust built over a decade. Memory is about cost-per-bit and granular demand from hyperscalers.
Samsung is the only major player who fights on both fronts. TSMC only does foundry. SK Hynix only does memory (mostly HBM now). Samsung has to allocate R&D, capital, and executive brainspace to both. This is an asset, but only if you win both. If you lose one, the other subsidizes a corpse.
Historically, foundry was a side-bet. Samsung could win DRAM cycles and plow 30% of revenue into CapEx for new fabs, but a portion of that always leaked into logic R&D. The bet was always: “We can design software (Exynos) and manufacture it ourselves. We are a self-contained IDM.” That bet is failing.
The Core: 85 Trillion is a Red Flag
Let's interrogate the 85 trillion. A 50% margin means the price of DRAM and NAND has essentially doubled from the cycle low. That is entirely possible given the HBM ramp. The ASP of an HBM3E stack is 8-10x a standard DDR5 DIMM. Samsung has volume.
But here is the math I care about: the internal transfer price.
Samsung’s foundry division (DS) buys wafers from Samsung's memory division. When memory is hot, the wafer price increases. This artificially inflates the cost base for the foundry division. Meanwhile, the foundry division's external revenue is weak. It is losing logic customers to TSMC (Qualcomm, AMD, NVIDIA). Its internal customer (Exynos) is a marginal player.
The result: The 85 trillion profit is almost entirely from the memory division's external sales. The foundry division's operating profit is likely deeply negative, possibly a loss of KRW 5-10 trillion this quarter. The memory division is not just funding the parent company's dividends. It is covering the foundry division's fixed costs, equipment depreciation, and R&D burn.
The market sees headline profit. I see a single-point-of-failure risk. The entire Samsung narrative is currently pegged to a single variable: the HBM supply-demand balance. If that balance shifts even 5% (new capacity from SK Hynix comes online, a hyperscaler reduces orders), the margin disappears. The foundry losses become visible.
This is not a growth story. This is a risk-on trade with a massive, hidden short on the foundry division.
The Contrarian: The "Integration" Thesis is Wrong
The bullish take says: "Samsung is the only one who can make HBM4 with its own logic interface die and its own advanced packaging." This is the "Integrated Advantage" playbook.
It is wrong for three reasons:
- TSMC owns the logic interface. For HBM4, the logic die (the base interface) is a critical component. SK Hynix is using TSMC for this. Samsung could use its own, but that logic die will be made on its 2nm process, which has zero volume credibility. The yield is unproven. If the interface die has a 20% defect rate, the entire HBM stack is scrap.
- You can't compete on speed and scale simultaneously. Samsung wants to sell HBM to NVIDIA. NVIDIA needs the highest performance and the tightest delivery. TSMC can deliver the logic die quickly because its 2nm tooling is months ahead. SK Hynix can deliver the memory because it is slaved to HBM production. Samsung has to manage both pipelines. A delay in one creates a bottleneck in the other.
- The geopolitical dimension is a liability. The US desperately wants a "Trusted Foundry" alternative to TSMC. But Samsung is a Korean company with massive exposure to China (NAND fab in Xi'an). If the US sanctions China further over chips, Samsung is caught in the middle. Its American fab in Taylor, Texas, is supposed to be a safe harbor, but it is two years behind schedule.
This is the classic CEO trap: "We must be everything to everyone." It sounds good on a conference call. It ruins your margin structure in reality.
The Takeaway: This is a Short Opportunity on the Narrative
I am not shorting Samsung stock today. The momentum is strong. The Q2 print will be a blowout.
But I am watching for the catalyst. The catalyst is the SK Hynix HBM4 volume ramp in late 2025. If SK Hynix demonstrates they can deliver HBM4 at scale using TSMC logic, the "Samsung Integration" thesis dies. The market will realize that Samsung's foundry is a strategic liability, not an asset.
The price action will mirror what happened to Intel in 2020-2022: a great product (memory for Samsung, CPU for Intel) hides a foundational structural decay. When the product cycle turns, the decay is exposed. The dividend gets cut. The CEO gets replaced.
**The play: Wait for the HBM4 news to break. If Samsung delays its 2nm interface die by one quarter, open a medium-term short position. The profit illusion will evaporate."
— The only way to fix this is to break the company up. Memory and Foundry must be separated. But the founder's family won't do that. So the trader just waits.
