The Current Case for Memory Stocks

Key Takeaways

• AI workloads are turning memory from a cyclical commodity into a differentiated technology product.

• High barriers to entry in DRAM and HBM, plus a more disciplined industry, support pricing power and margins.

• Memory stocks are still priced as cyclicals, which understates AI’s structural impact and their earnings potential.

Despite a sharp drawdown over the summer, memory chip makers have been among the best-performing companies connected to the buildout of artificial intelligence (AI) infrastructure. We believe recent innovations hastened by the workload requirements of large language models (LLMs) have shifted memory from a cyclical commodity into a differentiated technology product with attractive longer-term risk reward characteristics.

AI highlights the importance of memory demand. Hyperscalers and private LLMs are moving beyond training toward inference and agentic AI, where systems must retain greater context and handle increasingly complex tasks. As a result, memory is becoming a key determinant of overall AI performance, improving the productivity of existing compute infrastructure rather than simply supporting it.

At the same time, memory supply remains structurally constrained. Leading-edge direct random access memory (DRAM) and high-bandwidth memory (HBM) require advanced manufacturing, complex packaging and years of accumulated expertise. High barriers to entry limit the industry’s ability to rapidly add capacity, reducing the risk of the severe oversupply cycles that have historically characterized memory markets. Chinese tech companies are several years behind on HBM, held back by trade restrictions on important equipment such as extreme ultraviolet lithography machines.

As a result, HBM is transforming from commodity-like memory chips into critical, differentiated technologies enabling AI workloads similar to GPU, CPUs and other logic chips (Exhibit 1). Competitive advantage increasingly depends on engineering capability, packaging expertise and customer qualification rather than pure scale. This supports stronger pricing power, higher margins and improved returns on capital for industry leaders.

Exhibit 1: Memory Demand Shifting to AI Applications

As of September 30, 2026. Source: ClearBridge Investments and company reports.

Moreover, industry economics are improving structurally. Memory is more consolidated, especially in HBM, more supply-disciplined and increasingly exposed to higher-value AI products than in prior cycles. If these changes persist, earnings and profitability should prove materially more resilient than in past cycles.

Despite significant share price appreciation, accelerating earnings have kept memory valuations attractive. Following the recent correction, many memory stocks continue to be valued as cyclical businesses vulnerable to future oversupply. In our view, this underappreciates the structural impact of AI, the growing importance of HBM and the improving quality of industry earnings. If memory is evolving from a commodity market into a critical enabler of AI infrastructure, current valuations appear undemanding relative to their earnings potential.