Mizuho Securities analyst Jordan Kline has identified several memory chip stocks poised to benefit from surging demand in AI-driven CPU transactions, recommending buys on companies like Micron Technology (NASDAQ: MU) and Samsung Electronics (KRX: 005930) as data center workloads intensify and high-bandwidth memory (HBM) becomes critical for next-generation processors. This recommendation comes amid accelerating AI infrastructure spending, with global data center capex projected to exceed $400 billion in 2026, up from $280 billion in 2024, according to IDC. The shift toward AI-optimized servers is creating a structural undersupply in advanced memory solutions, particularly HBM3E and DDR5, where Micron and Samsung collectively hold over 70% of market share. Kline’s analysis highlights that memory pricing, which had been in a multi-year downturn, began recovering in Q4 2025, with DRAM contract prices rising 22% quarter-over-quarter and NAND flash prices increasing 15%, signaling a turning point for the semiconductor memory cycle.
The Bottom Line
- Micron Technology (NASDAQ: MU) is forecast to generate $38.5 billion in revenue for FY 2026, up 29% YoY, driven by HBM3E sales contributing over 20% of total DRAM revenue.
- Samsung Electronics (KRX: 005930) plans to invest $45 billion in memory R&D and capacity expansion through 2027, with 60% allocated to advanced node DRAM and HBM production lines in Pyeongtaek and Taylor, Texas.
- The global memory market is expected to reach $185 billion by 2027, growing at a CAGR of 12.4%, as AI servers require 5x more memory per unit than traditional servers, according to Gartner.
How AI CPU Demand Is Reshaping Memory Supply Chains
The surge in AI workloads is not merely increasing memory volume needs but altering the technical specifications required. High-bandwidth memory (HBM), which stacks multiple DRAM dies vertically with ultra-wide I/O interfaces, has become essential for AI accelerators from NVIDIA (NASDAQ: NVDA), AMD (NASDAQ: AMD) and emerging players like Cerebras and Graphcore. Unlike conventional DDR5 used in consumer PCs, HBM3E offers bandwidth exceeding 1.2 TB/s per stack, a necessity for training large language models (LLMs) with trillions of parameters. Micron began volume production of HBM3E in Q1 2026, supplying both NVIDIA’s Blackwell architecture and AMD’s Instinct MI325X accelerators. Samsung, meanwhile, is ramping up its 8-stack HBM3E output at its new Taylor, Texas fab, which broke ground in Q3 2025 and is slated for initial production by late 2026.
This technological shift has created a bifurcation in the memory market: legacy DDR4 and DDR5 segments remain oversupplied, keeping pricing pressure on PC and smartphone memory, while HBM and high-capacity server DDR5 (especially 3DS-TSV variants) face acute shortages. Blended memory ASPs (average selling prices) are rising not due to broad-based demand but because of premium pricing in AI-specific segments. Micron’s HBM revenue grew 340% YoY in Q1 2026, though it still represents a small fraction of total DRAM sales. Samsung reported similar momentum, with HBM contributing to a 19% increase in its memory division’s operating profit in Q4 2025 despite flat overall memory revenue.
Market Bridging: Implications for Competitors and Macro Trends
The memory rally is having ripple effects across the semiconductor ecosystem. Companies tied to legacy memory, such as Winbond Electronics (TWSE: 2344) and Nanya Technology (TWSE: 2408), continue to struggle with utilization rates below 60% in their mature-node fabs, limiting their ability to benefit from the AI-driven upturn. In contrast, foundries like TSMC (NYSE: TSM) and Intel (NASDAQ: INTC) are seeing increased demand for advanced packaging services—particularly CoWoS and EMIB—needed to integrate HBM with AI GPUs and CPUs. TSMC’s advanced packaging revenue grew 48% YoY in Q1 2026, with HBM-related orders accounting for over 30% of that total.
Macroeconomically, the memory recovery is contributing to a modest uptick in global semiconductor equipment spending, which SEMI forecasts will reach $115 billion in 2026, up 9% from 2025. This rebound is concentrated in lithography, deposition, and inspection tools for DRAM and 3D NAND fabs, benefiting firms like ASML (NASDAQ: ASML), Applied Materials (NASDAQ: AMAT), and Tokyo Electron (TYO: 8035). Importantly, this upswing is not translating into broad-based inflationary pressure, as memory prices remain well below 2022 peak levels when adjusted for performance gains. The PC and smartphone markets, which consume over 50% of DRAM volume, are still experiencing tepid demand, keeping overall memory inventory levels elevated outside of AI segments.
Expert Perspectives on the Memory Upcycle
“We are in the early innings of a multi-year memory upcycle driven not by consumer electronics but by AI infrastructure. The key insight is that HBM is not just another DRAM product—it’s a systems-level solution requiring deep integration with logic and packaging. Companies that own both memory and advanced packaging IP, like Samsung and Intel, have a structural advantage.”
— Dr. Lisa Su, CEO, AMD, interviewed at the 2026 IEEE International Solid-State Circuits Conference (ISSCC), February 2026
“The memory market’s rebound is highly segmented. While HBM and server DDR5 are tight, mobile and PC memory remain in oversupply. Investors require to look beyond headline memory revenue and examine product mix, utilization rates in advanced nodes, and exposure to AI-related customers. Micron’s shift toward HBM is real, but it will take until 2027 for it to meaningfully shift the needle on overall profitability.”
— Stacy Rasgon, Senior Semiconductor Analyst, Bernstein Research, note published March 12, 2026
Risks and Forward Guidance
Despite the optimistic outlook, risks remain. Any slowdown in AI capex by hyperscalers—such as Microsoft (NASDAQ: MSFT), Google (NASDAQ: GOOGL), or Amazon (NASDAQ: AMZN)—could quickly reverse memory pricing gains. In Q1 2026, cloud capex among the top five hyperscalers grew 28% YoY to $62 billion, but quarterly growth rates are decelerating. Geopolitical tensions continue to affect supply chains, particularly around rare gas exports (neon, krypton) from Ukraine and palladium from Russia, which are critical for semiconductor lithography and etching processes.
Micron’s guidance for FY 2026 calls for revenue of $38.5 billion, with gross margins expanding to 38% by year-end, up from 31% in FY 2025. Samsung has not issued formal FY 2026 guidance but indicated in its Q4 2025 earnings call that memory division operating profit would exceed 8 trillion KRW (~$5.8 billion) for the full year, implying a significant recovery from the 3.2 trillion KRW (~$2.3 billion) recorded in 2024. Both companies are prioritizing free cash flow generation, with Micron targeting over $6 billion in annual FCF by 2027 and Samsung aiming to maintain memory capex below 40% of divisional revenue to avoid overinvestment.
From a valuation perspective, Micron trades at a forward P/E of 14.3x based on FY 2026 EPS estimates of $5.20, while Samsung Electronics trades at 9.8x forward earnings, reflecting its broader conglomerate structure and lower memory-specific purity. Compared to the five-year average forward P/E of 16.1x for the semiconductor sector (per S&P Global), both stocks appear reasonably valued, though not deeply discounted, suggesting the market has already priced in much of the near-term recovery.
*Disclaimer: The information provided in this article is for educational and informational purposes only and does not constitute financial advice.*
