AI Reshapes the Memory Landscape: Samsung's Profit Ascent and a Shift in Investment Logic

A new Morgan Stanley industry report delves into the seismic shifts within the semiconductor sector, driven overwhelmingly by artificial intelligence. The core thesis is that AI data center demand is propelling the memory industry into a structural upturn, marked by long-term contracts and firmer pricing floors that have altered its risk profile.

The New Profit King: Samsung Set to Overtake NVIDIA

The report makes a striking forecast: by the third quarter of 2026, Samsung Electronics is projected to post a quarterly operating profit of approximately $81.3 billion. This figure would surpass an estimated $71.6 billion for NVIDIA in the same period, potentially crowning Samsung as the world's most profitable company, albeit temporarily.

Despite this robust profit outlook, memory stock valuations tell a different story. Shares of Samsung and a key peer are down about 35% from their peaks, trading at forward P/E ratios of just 3-4x. Morgan Stanley argues this indicates the market is still pricing these stocks within the old cyclical framework, underestimating the sector's structural improvements.

HBM Market Boom, But Cyclical Risks Linger

High-Bandwidth Memory (HBM), the essential companion to AI chips, is at the heart of this growth surge. The HBM market is expected to triple in size to around $100 billion by 2028. Demand is so fierce that industry capacity has been sold out years in advance, largely under long-term agreements, making a supply glut unlikely before 2027.

  • Global View: The memory sector is in an AI-fueled "super cycle."
  • China's Role: Chinese memory makers now hold an estimated 12-13% share of the global NAND flash market, with their HBM products capturing about 20% of the domestic market.

However, the report sounds a note of caution: a "super cycle" does not mean the traditional cycle is dead. The fundamental economics of the industry dictate that new supply will eventually arrive, attracted by high profitability. While the risk profile has changed, the cyclical nature has not been erased.

Investment Stance: Overweight Compute, Underweight Memory

Based on this analysis, Morgan Stanley's portfolio advice is clear: investors should be relatively underweight memory stocks while being overweight AI compute and networking infrastructure. Within the compute segment, top picks include NVIDIA, TSMC, and Broadcom.

Near-term fundamentals remain strong. The report anticipates DRAM contract prices to rise 10-15% quarter-over-quarter in Q4, with NAND flash prices increasing 15-20%. This provides solid short-term earnings support for related firms.

The key takeaway is that investors must adopt a more nuanced view of the semiconductor boom. AI presents a historic opportunity, but the investment logic and risk-reward profile are diverging sharply across different sub-sectors.