SK Hynix released its Q1 2026 earnings, posting operating profit of 37.61 trillion won — slightly below the 37.92 trillion consensus but barely noticed. What caught investors' eyes was the 72% operating margin, the highest quarterly figure ever recorded in the semiconductor industry.
The growth trajectory is even more staggering: profit surged fivefold year-over-year and nearly doubled quarter-over-quarter. Such exponential curves were once confined to startup pitch decks; now they belong to a legacy memory manufacturer with tens of thousands of employees and multibillion-dollar fabrication plants.
AI Inference Rewrites Memory Demand Equation
HBM (High Bandwidth Memory) vertically stacks multiple DRAM layers, allowing AI accelerators to read and write far more data per clock cycle — no more queuing. During AI training, matrix multiplication dominates and can be parallelized. But during inference, each “answer” requires frequent access to different memory addresses, making HBM's bandwidth advantage irreplaceable.
SK Hynix stated in its report that AI is transitioning from “large model training” to an Agentic AI phase. Every inference step triggers a memory read/write event. As AI evolves from one-off conversations to persistent digital agents, the demand baseline for DRAM and NAND shifts from “occasional spikes” to “continuous floor.”
Reports indicate that SK Hynix has pre-sold nearly all of its DRAM, NAND, and HBM capacity through end-2026 to NVIDIA.
Valuation Paradox: Soaring Profits Yet PE Below 10x
Samsung and SK Hynix are expected to see net profit surges of 400% and nearly 300% this year, far exceeding TSMC's roughly 50%. Samsung's full-year net profit could reach $151 billion, SK Hynix $115 billion — both surpassing TSMC's $81 billion.
Yet the valuation gap is stark: NVIDIA trades at 22x PE, while memory stocks like Micron and Kioxia carry forward PEs below 10x, SK Hynix included. The market is pricing memory as cyclical plays — great earnings now, but a reversal could come anytime.
Whether memory has truly transformed from a “commodity cycle” into “AI infrastructure” will only be known at the next downturn. But for this quarter, 72% operating margin answers a more straightforward question: who is the most definitive beneficiary of the AI era.