Chip Industry
Samsung's profit surge cannot hide industry concerns: the tug-of-war between memory recovery signals and the AI cycle inflection point.
Samsung Electronics' profit surged year-on-year in the second quarter of 2025, but chip stocks such as Micron and Nvidia collectively declined, as the market experienced intense contention between signals of memory recovery and expectations of slowing AI demand. This article provides an in-depth analysis of this contradictory phenomenon from the perspectives of the industrial chain, technology roadmap, competitive landscape, and regional supply chain.
Event: Divergence Between Profit and Stock Price In July 2025, Samsung Electronics released its preliminary earnings report for the second quarter of 2025, with operating profit surging over 200% year-on-year, hitting a three-year high. This performance was mainly driven by explosive growth in demand for HBM (High Bandwidth Memory) and the continued recovery of DRAM/NAND prices. However, in stark contrast to Samsung's impressive financial data, Micron Technology's stock price fell over 4% on the same day, while AI chip leaders such as Nvidia and AMD also experienced varying degrees of pullback. This divergence has sparked widespread market discussion about a potential turning point in the semiconductor industry cycle.
Background: Memory Recovery and AI Driving Forces ### Corporate Background Samsung Electronics is the world's largest memory chip manufacturer, holding approximately 40% and 35% market shares in DRAM and NAND, respectively. Its profit growth directly reflects the recovery of the memory market from the deep recession in 2023. Micron, as the largest U.S. memory chip maker, also benefits from the price hike cycle, but investors are divided on its future growth expectations.
Technology Background The core driving force behind this memory recovery is the huge demand for HBM from AI servers. HBM3e has become the standard configuration for Nvidia's H200/B200 GPUs, with a single GPU requiring multiple HBM chips, significantly increasing DRAM bit consumption. Meanwhile, DRAM demand for general servers and PCs remains weak.
Market Background In the first half of 2025, the semiconductor market showed a clear "ice and fire" situation: AI-related chips (GPU, HBM, AI ASIC) were in short supply, while inventories for consumer electronics, automotive, and industrial chips remained high. The market worries that AI capital expenditure growth may slow down in 2026, leading to a synchronized cooling of memory and GPU demand.
In-depth Analysis
Technology Impact - HBM Technology Roadmap: HBM4 is expected to enter mass production in 2026, introducing hybrid bonding and more advanced packaging processes. Samsung, SK Hynix, and Micron are racing to increase stack layers and bandwidth. The technical barriers lie in TSV (Through Silicon Via) yield and thermal management. - DRAM Process: 1b nm (approximately 12nm) DRAM has become mainstream. Samsung and SK Hynix have begun transitioning to 1c nm, while Micron has fallen slightly behind after the 1β nm node. - NAND Architecture: 3D NAND with over 300 layers has entered mass production, with competition intensifying between Samsung's V-NAND and YMTC's Xtacking technology.### Supply Chain Impact - Storage supply chain: Upstream equipment makers (ASML, Lam Research, Tokyo Electron) benefit from DRAM fab expansion; materials suppliers (silicon wafers, photoresists) see stable demand. However, if HBM demand falls short, the risk of DRAM overcapacity will re-emerge. - AI supply chain: HBM packaging players (ASE, Amkor) and test equipment makers (Advantest, Teradyne) face order fluctuations. Server OEMs (Dell, Supermicro) inventory adjustments could feed through to chip procurement. - Risk factors: Rising inventory levels and shorter lead times are leading indicators of weakening demand.
Competitive Landscape - Divergence among the three storage giants: Samsung leverages its first-mover advantage in HBM3e to consolidate its leading position; SK Hynix closely follows in share with major clients (NVIDIA); Micron lags in HBM and is more reliant on traditional DRAM, making its stock price more sensitive to the cycle. - AI chip competition: Demand for NVIDIA H200/B200 remains strong, but AMD MI350, Intel Gaudi 3, and custom chips (Google TPU v6, Amazon Trainium 3) are beginning to erode share. The AI chip market is transitioning from a single hegemon to multiple strong players. - Wafer foundry: Samsung's 3nm GAA yield improves but customer acquisition is slow; TSMC's 3nm remains fully loaded, with 2nm expected to trial produce in the second half of 2025. Advanced process capacity remains concentrated at TSMC.
Regional Implications - United States: Micron's stock decline reflects fragile AI investment sentiment in US equities; fabs funded by the CHIPS Act (Micron in New York and Idaho) have long construction cycles and cannot change the supply chain in the short term. - China: Domestic storage (YMTC, CXMT) has made progress in mature nodes, but advanced products like HBM remain subject to export controls. The demand for AI chip self-sufficiency is driving indigenous HBM R&D. - South Korea: Samsung and SK Hynix's profits rely on HBM, but if AI demand slows, South Korea's storage industry will be the first to suffer. The government is increasing semiconductor supply chain subsidies. - Japan: Japan benefits from materials (Shin-Etsu Chemical, JSR) and equipment (Tokyo Electron). Rapidus plans to mass-produce 2nm in 2027, but commercialization is still far off. - Taiwan, China: TSMC's advanced packaging capacity is fully loaded; the expansion speed of CoWoS capacity determines AI chip shipment volume. Memory packaging and testing players (Powertech, ChipMOS) are driven by HBM packaging demand. - Europe: Infineon and NXP are still bogged down by automotive chip inventory digestion, with limited impact on memory demand.### Investment Perspective - Market Sentiment: Samsung's profit surged but its stock price did not rise significantly, indicating that the market has already priced in the memory recovery. Micron's decline reflects skepticism about the sustainability of bit shipments and average selling prices (ASP) in the second half of the year. - Valuation: Micron's current P/E ratio is around 20x, close to its historical median, but if the memory cycle peaks, valuations could compress again. NVIDIA's P/E ratio exceeds 40x, requiring continuous earnings beats to support it. - Capital Expenditure: Samsung's 2025 capital expenditure plan is approximately $40 billion, mostly allocated to memory and foundry operations. If demand weakens, a reduction in capital spending will impact equipment stocks.
Long-Term Outlook - Next 3 Years: The HBM market will maintain a compound annual growth rate of over 30%, but traditional memory demand will only grow by single digits. Memory manufacturers need to balance HBM and general DRAM capacity. - Next 5 Years: New technologies such as in-memory computing and CXL memory may reshape the memory hierarchy, transitioning memory from a "supporting role" to a "key computing component." - Next 10 Years: The AI general computing era may give rise to new memory architectures, such as MRAM and phase-change memory, but NAND/DRAM will still dominate in the short term.
Industry Chain Analysis Upstream: Equipment (ASML, Lam Research) order visibility is affected by the pace of memory fab expansion; materials (GlobalWafers, Showa Denko) benefit from shipment growth, but price elasticity is limited.
Midstream: Memory fabs (Samsung, SK Hynix, Micron) compete on HBM yield and next-generation DRAM processes; foundries focus on advanced logic process nodes, but are indirectly related to memory demand.
Downstream: AI servers (NVIDIA, AMD) and cloud service providers (Microsoft, Google, Amazon) are the ultimate buyers of HBM; PC/smartphone OEMs (Apple, Samsung Electronics) inventory adjustments impact NAND and LPDDR demand.
Conclusion The divergence between Samsung's soaring profits and falling chip stocks fundamentally reflects the market's skepticism about the sustainability of the semiconductor cyclical recovery. AI drives record-high memory demand, but the drag from traditional applications remains evident. Investors are shifting from "scrambling for supply" to "pricing in the cycle peak." Over the medium to long term, only companies with a technological moat (advanced HBM packaging, leading DRAM processes) and a diversified customer base (beyond AI) can navigate the cycle. All segments of the industry chain should be wary of the inventory correction that may occur from the second half of 2025 to 2026.
Desk context · semiconreport
semiconreport frames this note through Semicon Report tracks chip design, fabrication, AI compute demand, supply-chain shifts, market cycles, and.... dates, names and status changes still need checking: Source links should be opened before the summary is reused. Chip Industry / Industry brief / Focus explains the local editorial angle.