Samsung Electronics and SK Hynix Q3 Earnings Approaching: AI Memory Super Cycle Resilience Under Scrutiny
nashnova research
Samsung Electronics and SK Hynix are on track for a combined Q3 operating profit approaching the historic ₩189.9 trillion mark, yet analysts have quietly trimmed forecasts over the past three months — putting the AI-driven memory super-cycle through its first real stress test.
The numbers look stellar — so why are analysts cutting estimates?
FnGuide consensus puts Samsung's Q3 revenue at ₩199.1 trillion with ₩105.6 trillion operating profit; SK Hynix at ₩94.1 trillion revenue and ₩74.1 trillion operating profit.
Over the past three months, analysts cut Samsung's revenue and profit estimates by 2.6% and 4.4% respectively; SK Hynix by 5.3% and 5%.
This means → the downgrades stem from a stronger Korean won shrinking dollar-denominated sales, not from a weakening memory market. In plain terms = they sold the same volumes, but each dollar earned converts into fewer won on the books.
Memory prices are still rising — so why is the momentum fading?
Sharp Q2 price increases in DRAM and NAND flash were the main engine behind both companies' strong results. Prices are expected to keep climbing in Q3, but more slowly — Mirae Asset Securities estimates Samsung's Q3 DRAM ASP will rise 16.5%, then narrow to 5.4% in Q4.
Long-term agreements (LTAs) — contracts that lock in future capacity and pricing — are part of the reason. Samsung has tied up 60%–70% of its memory capacity through LTAs; SK Hynix completed LTA negotiations with roughly ten key customers in Q2.
This means → LTAs provide revenue stability, but during a rapid price upcycle, contract prices lag spot prices — capping some of the upside on margins.
Why is HBM4 the single biggest thing to watch in these reports?
Samsung expects Q3 HBM4 — the fourth generation of high-bandwidth memory, ultra-fast RAM designed specifically for AI chips — sales to more than triple quarter-on-quarter, accounting for over 60% of its total HBM revenue in H2 2026.
SK Hynix faces a two-front challenge: defending its market lead in HBM3E while simultaneously scaling HBM4 production.
This reflects a deeper shift in AI infrastructure — the bottleneck is moving from "who has the GPUs" to "whose memory can keep up." As a KB Securities analyst put it: "The GPU is the heart; memory is the circulatory system."
Agentic AI is here — why does that make memory matter even more?
According to a KB Securities report, Meta's launch of its agent-based AI application Muse signals a new phase that demands far higher memory bandwidth.
Traditional generative AI processes roughly 100 tokens per second; agentic AI ramps that to 1,000 tokens per second — a tenfold jump in data throughput.
In plain terms = older AI answers one question at a time; agentic AI runs ten tasks for you simultaneously, demanding an entirely different class of memory bandwidth. This means → the performance bottleneck in data centers is shifting from compute to storage.
Foundry turnaround vs. the "chip-inflation paradox" — what is Samsung's internal tension?
Samsung's foundry division has been loss-making since 2023. Boosted by expanded 4 nm capacity and improved yields, it is widely expected to swing to profit in Q3.
Yet the device-experience division — smartphones and TVs — is caught in a "chip-inflation paradox": strong Q2 sales of premium Galaxy foldables still produced an ₩8 billion operating loss because component costs rose faster than revenue.
In plain terms = rising memory prices are great news for the chip division but become a cost burden for the handset division — what the left hand earns, the right hand pays for.
What are these earnings reports really testing?
The core suspense: if HBM4 profit contribution falls short or DRAM and NAND price growth slows faster than expected, the debate over whether semiconductors have already "peaked" will reignite.
Samsung is already looking beyond HBM4 to zHBM — stacking memory directly on top of GPUs — and aims to deliver samples by late 2027.
This means → these two Q3 reports are not just an earnings release; they are the market's key checkpoint for whether the AI memory super-cycle is a structural trend or a cyclical peak.
市场有风险,内容仅供研究参考,不构成投资建议。
