SK Hynix Q1 operating margin reached 72%, setting a record for the highest single-quarter figure in semiconductor industry history.

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯1060 words

SK Hynix's operating profit margin reached 72% in Q1 2026, setting an all-time record for the semiconductor industry. The driving force behind this is singular: the demand for High Bandwidth Memory (HBM) for AI inference has exceeded everyone's forecasts. (Context: DRAM is set to skyrocket! Samsung strike vote concludes; is it better to buy memory than gold now?) (Background: South Korean youth have stopped trading crypto and are flocking to Samsung and SK Hynix! Replicating the "TSMC faith" of the Taiwanese?) SK Hynix released its Q1 earnings report, showing an operating profit of 37.61 trillion KRW. While slightly lower than the market expectation of 37.92 trillion KRW, almost no one cared. The figure investors were truly focused on was the 72% operating profit margin, the highest in the company's history, breaking records in the semiconductor industry. Even more staggering is the growth rate: a five-fold year-on-year increase and nearly a doubling quarter-on-quarter. This type of exponential curve, previously seen only in the PPTs of startups, has now appeared in the financial report of a veteran memory manufacturer with tens of thousands of employees and factories costing over 10 billion USD. AI inference has rewritten the demand equation for memory. HBM stands for High Bandwidth Memory. It vertically stacks multiple layers of DRAM chips, allowing AI accelerators to read and write massive amounts of data within a single clock cycle, rather than queuing for transmission one "lane" at a time. During AI model training, massive computations are concentrated on matrix multiplication, which can be compressed through parallelization. However, every time a model performs inference to "answer a question," it requires frequent access to different memory addresses, making the bandwidth advantage of HBM irreplaceable. SK Hynix's earnings report notes that AI is transitioning from "large model training" to the "Agentic AI" stage. Every inference step is a memory read/write event. As AI evolves from single conversations to continuously running digital agents, the demand base for DRAM and NAND has shifted from "occasional peaks" to a "sustained baseline." Reportedly, SK Hynix has already pre-sold almost all of its DRAM, NAND, and HBM production capacity through the end of 2026 to NVIDIA. The valuation paradox remains unanswered. Net profits for Samsung and SK Hynix are expected to surge by 400% and nearly 300% respectively this year, far exceeding TSMC's growth of approximately 50%. Samsung's annual net profit is projected to reach 151 billion USD, and SK Hynix's 115 billion USD, both surpassing TSMC's 81 billion USD. However, the valuation gap is clear: NVIDIA's P/E ratio is 22x, while memory stocks like Micron and Kioxia have forward P/E ratios of less than 10x, with SK Hynix roughly the same. The market is telling investors one thing through these valuations: even though stock prices have quadrupled over the past year, memory is still priced as a "cyclical stock"—profits are high for now, but a reversal could happen at any time. Whether memory has truly transformed from a "cyclical commodity" into "AI infrastructure" may not be answered until the next industry downturn. But at least for this quarter, the 72% profit margin has answered a more direct question: who is the most certain beneficiary of the AI era.

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching0 items

No similar events found (requires more data samples or embedding search; currently MVP keyword matching)

Raw Information

ID:81d20ca355

Source:動區 BlockTempo

Published:2026-04-23 02:24:58

Category:zh_news · Export Category zh

Symbols:Unspecified

Community Votes:+0 / −0 · ⭐ 0 Important · 💬 0 Comments