Samsung Posts Massive Q2 Profit on Memory Demand
For AI practitioners, prolonged memory undersupply and sharply higher DRAM/NAND prices raise the marginal cost of large-scale inference and HBM-dependent workloads, affecting cloud and on-prem cost planning. Reuters reports that Samsung Electronics is likely to estimate its operating profit jumped about 18-fold to 86 trillion won for Q2, up from 4.7 trillion won a year earlier, based on an LSEG SmartEstimate of 30 analysts' forecasts. Reuters reports this would be Samsung's third consecutive quarter of record operating profit as AI-driven demand outpaces supply. Reuters reports Citi Research found average selling prices for DRAM and NAND rose 44% and 53% quarter-on-quarter, and Reuters reports analysts say agentic AI and inference workloads are widening memory demand. Reuters reports analysts flagged potential AI infrastructure delays as the largest near-term risk.
Editorial analysis
The headline matter for practitioners is cost and capacity. Sustained memory undersupply and large quarter-on-quarter price jumps for DRAM and NAND elevate the cost base of both training and inference, especially where HBM and large-memory servers are required. This changes procurement and optimisation priorities for teams managing large models and inference fleets.
What happened - Reuters reports that Samsung Electronics is likely to estimate operating profit rose about 18-fold to 86 trillion won for the April-June quarter, up from 4.7 trillion won a year earlier, according to an LSEG SmartEstimate based on forecasts from 30 analysts. Reuters reports this would mark a third consecutive quarter of record operating profit driven by a prolonged memory shortage as demand from AI inference infrastructure continues to outpace supply. Reuters reports analysts expect the memory market to remain undersupplied at least through next year.
Market signals and pricing - Reuters reports that Citi Research said average selling prices for DRAM and NAND climbed 44% and 53% quarter-on-quarter, respectively. Reuters reports analysts cited stronger demand not only for high-bandwidth memory (HBM) but also for conventional DRAM and NAND as AI, and specifically agentic AI workloads, expand into broader computing tasks that need more working memory and storage during inference.
Editorial analysis - technical context
Industry-pattern observations show that when hyperscaler and accelerator demand concentrates on HBM and large-memory server configurations, supply tightness propagates into broader DRAM and NAND markets. That typically raises spot and contracted prices, lengthens lead times, and forces engineering teams to prioritise memory efficiency, model quantisation, sharding, or storage-tiering strategies.
What to watch
- •Samsung's formal quarterly results to confirm the estimate and provide guidance, Reuters reports.
- •Capex and capacity plans from major memory makers and their timing.
- •HBM shipment trends and hyperscaler ordering patterns.
- •Cloud instance pricing and availability for memory-heavy instance types.
Key Points
- 1Sustained memory undersupply lifts DRAM/NAND prices, increasing the operational cost of large-scale inference and HBM-heavy workloads.
- 2Agentic AI and broader inference use are shifting demand from training-focused HBM to more general DRAM and NAND capacity needs.
- 3Practitioners should monitor supplier capacity, HBM shipment trends, and cloud memory-instance pricing as indicators of medium-term cost pressure.
Scoring Rationale
Rising memory prices and persistent undersupply materially affect AI infrastructure costs and capacity planning for practitioners; the story is notable but not a frontier-model release.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
