Samsung Electronics posted a first-quarter operating profit of 57.2 trillion Korean won, up over 750% year-over-year, beating analyst consensus of 55.28 trillion won. Revenue hit 133.9 trillion won, roughly 70% higher than the same period last year, setting a new quarterly record. The numbers align with Samsung’s own guidance and already exceed its full-year 2025 profit of 43.6 trillion won.

The primary driver: the Device Solutions division, which houses memory chip, semiconductor design, and foundry units. DS posted an operating profit of 53.7 trillion won—up from just 1 trillion won a year ago—on revenue of 81.7 trillion won, a 225% increase. The memory business alone “surpassed its quarterly sales record by addressing high-value-added AI demand despite limited supply availability,” per Samsung’s earnings statement.

This supply crunch isn’t accidental. Hyperscalers are absorbing available HBM and server DRAM capacity for AI workloads, pushing prices higher across the memory stack. Samsung noted that server memory demand should remain strong into the second half, driven by continued AI adoption and accelerating agentic AI use cases. The trade-off: consumer electronics memory faces tighter supply as fabs prioritize high-margin AI parts.

Samsung is expanding its high-bandwidth memory (HBM) portfolio—critical for Nvidia’s data center GPUs—but still trails SK Hynix in market share. Counterpoint Research puts SK Hynix at 57% of HBM revenue in Q4 2024. Samsung shipped its first HBM4 chips in February, roughly a year after SK Hynix delivered samples. Ray Wang of SemiAnalysis notes the gap has narrowed versus previous HBM generations, but SK Hynix retains the lead. For data engineers, the takeaway is clear: memory supply constraints are structural, not cyclical, and AI workload planning must account for continued allocation pressure through 2025.

Source: CNBC