🚀 It Wasn’t Just HBM: The Powerful Comeback of NAND in the AI Era

For the past two years, the semiconductor narrative has revolved almost entirely around HBM (High Bandwidth Memory). But in 2026, a new storyline is emerging—the unexpected resurgence of NAND flash.

SanDisk, widely known for USB drives and SD cards, delivered what can only be described as a “super surprise” earnings report:

• Operating profit surged 386%
• EPS beat expectations by 77%
• Stock jumped 17% in after-hours trading

This wasn’t just a strong quarter—it signaled a structural shift in AI infrastructure demand.

📊 Why Is NAND Suddenly Surging?

Until recently, AI demand was centered on training models. The focus was on:

• GPUs (NVIDIA)
• High-speed memory like HBM

Training required raw computational power and high-bandwidth data transfer.

However, the AI market is now rapidly shifting toward inference—real-world deployment and application of trained models.

In the inference phase, performance depends not only on computing speed but also on:

How fast massive amounts of data can be retrieved and processed.

This is where enterprise-grade SSDs (eSSD) powered by NAND flash become critical.

🧠 AI Inference Creates a Data Bottleneck

Even the fastest GPU becomes inefficient if storage cannot deliver data quickly enough.

In response:

• Ultra-high-capacity SSD demand is soaring
• AI servers are integrating thousands of SSD units
• GPU-to-SSD proximity reduces latency

NVIDIA’s next-generation AI accelerator, Vera Rubin, reportedly includes up to 1,152TB of SSD storage.

In simple terms:
AI infrastructure is evolving from a “GPU + HBM” model to a “GPU + HBM + Massive SSD” architecture.

📈 NAND Prices Soar, Market Turns Supplier-Favored

NAND flash export data shows explosive growth:

• 300% year-over-year increase in January
• HDD-to-SSD replacement accelerating
• Price gap narrowing from 5x to roughly 2x

As SSDs become more cost-effective relative to HDDs, enterprises are switching rapidly.

SanDisk has reportedly doubled prices and shifted to cash-first sales—clear signs of a supplier-driven market.

💡 ICMS Architecture: A New Memory Tier

NVIDIA introduced a concept called ICMS (Infrastructure Context Memory Storage).

ICMS represents a dedicated storage tier designed for AI’s long-term memory, often described as “cable cache.”

This new layer:

• Stores contextual AI data
• Optimizes inference workloads
• Elevates NAND’s strategic importance

If HBM serves short-term high-speed memory, NAND SSD now functions as AI’s long-term memory backbone.

🇰🇷 Why Korean Semiconductor Giants May Benefit Most

SanDisk ranks 5th globally in NAND market share.

Market leaders:

1. Samsung Electronics
2. SK Hynix

If the 5th-ranked company posts such explosive results, the implications for top-tier producers could be even larger.

In 2023–2024, HBM drove earnings momentum.

Now, NAND may provide a second wave of profit expansion for major semiconductor players.

📌 Investment Considerations

SanDisk’s stock has surged nearly 1,100% over six months, reflecting aggressive re-rating.

Key considerations:

• Is demand sustainable?
• Is inference-driven storage structurally permanent?
• How diversified should exposure be?

Rather than chasing momentum, investors may consider:

• Gradual accumulation on pullbacks
• Diversifying into semiconductor equipment and materials companies
• Monitoring NAND pricing trends closely

If AI inference growth continues, NAND demand could represent a multi-year structural tailwind.