NAND

NAND flash is non-volatile memory that stores data as trapped electrical charge, retaining it without power. It is the storage inside SSDs, phones, and AI data-center servers. Four firms—Samsung, SK Hynix, Kioxia, and Micron, plus SanDisk—make almost all of it, and AI demand has driven a supply crunch.

What NAND is, in plain terms

NAND is a type of **flash memory**—non-volatile storage that keeps data even when the power is off. It is what lives inside solid-state drives (SSDs), USB sticks, memory cards, and the storage chips in your phone and laptop. The name comes from the way its memory cells are wired together, resembling a 'NAND' logic gate. NAND is prized for high density and low cost per bit, which is why it replaced spinning hard drives in most devices. It is distinct from **DRAM**, the fast but volatile working memory that loses everything when power is cut.

How it works

Each NAND cell is a special transistor that stores data as a tiny amount of electrical charge, held either on a **floating gate** (a conductor) or in a **charge-trap layer** (an insulator). Adding or removing charge—via quantum tunneling—shifts the transistor's threshold voltage, and the chip reads that voltage level back as bits. Because the charge stays put without power, the data is non-volatile. The number of distinct voltage levels per cell sets the type: **SLC** stores 1 bit (2 levels), **MLC** 2 bits, **TLC** 3 bits, and **QLC** 4 bits (16 levels). More bits per cell means cheaper, denser storage but tighter voltage margins and shorter endurance, so controllers use error correction to keep data reliable.

From flat chips to 3D skyscrapers

For decades, makers shrank cells on a flat 2D plane, but that hit physical limits. The industry pivoted to **3D NAND** (Samsung brands it V-NAND), stacking memory cells vertically like floors in a skyscraper. In 2025, 3D NAND made up roughly 87% of the market. Layer counts have climbed past **300 layers**—SK Hynix shipped a 321-layer generation, and Samsung began mass-producing 1-terabit, 9th-generation QLC V-NAND in 2024. To go higher, makers fabricate separate stacks (e.g., 120 + 110 + 91 layers) and bond them together. Roadmaps point toward 500+ layers within a few years and a long-term goal near 1,000 by 2030.

Why it matters for AI and data centers

AI has turned NAND from a commodity into a bottleneck. Training and serving large models requires vast, fast storage for datasets, model checkpoints, and retrieval databases, driving a surge in **enterprise SSDs**—demand for which is projected to grow around 41%. Hyperscaler capital spending pulled NAND into a multi-quarter boom: high-density products are effectively sold out months ahead, with prices in early 2026 rising 55–60% quarter-over-quarter. Because wafer capacity is being reallocated toward AI memory and new fabs won't arrive in volume before late 2027, analysts describe a structural shortage rather than a normal cyclical dip.

Where NAND sits in the supply chain

NAND chips are made on silicon wafers in fabs by a handful of integrated device manufacturers, then packaged with a **flash controller** and firmware into finished SSDs and storage modules. This is separate from the silicon-photonics and optical-interconnect layer that moves data *between* AI accelerators—NAND is the persistent storage tier, sitting below DRAM and HBM in the memory hierarchy. The same advanced lithography and packaging ecosystem that supports logic and DRAM chips underpins NAND, which is why supply constraints ripple across smartphones, PCs, and servers at once.

Who the key players are

The market is highly concentrated. In Q3 2025, **Samsung** led with about 32% share, followed by **SK Hynix** (~19%), **Kioxia** (~15%, the former Toshiba memory unit), **Micron** (MU), and **SanDisk** (SNDK, ~12%). SanDisk became a pure-play NAND and SSD company in February 2025 when **Western Digital** spun off its flash business; that gives SNDK more direct NAND exposure but more cyclical risk than diversified Micron, which also makes DRAM and HBM. Together these firms control over 80% of global NAND supply.

Frequently asked

What does NAND stand for?

NAND refers to the 'NOT-AND' logic gate; the memory got the name because its cells are connected in series in a pattern that electrically resembles a NAND gate. The contrasting flash type, NOR flash, is wired differently and is used for code storage where random read access matters more.

What is the difference between NAND and DRAM?

NAND is non-volatile storage—it keeps data without power and is used for SSDs and phone storage. DRAM is volatile working memory that is much faster but loses all data when powered off. Most systems use both: DRAM as active memory and NAND as persistent storage.

Why are NAND and storage prices rising in 2025–2026?

AI data centers are consuming enormous amounts of memory and storage. Enterprise SSD demand is surging while fab capacity is being shifted toward AI memory, so high-density NAND is effectively sold out months ahead, pushing contract prices up sharply and likely keeping them elevated into 2027.

What is 3D NAND or V-NAND?

3D NAND stacks memory cells vertically instead of spreading them across a flat chip, dramatically increasing capacity per square millimeter. Samsung calls its version V-NAND. Leading 2025 chips exceed 300 stacked layers, with roadmaps aiming for 500-plus.

Which companies make NAND flash?

Samsung, SK Hynix, Kioxia, Micron, and SanDisk make almost all of the world's NAND, together holding over 80% of the market. Samsung is the leader; SanDisk became a standalone NAND company after its 2025 spin-off from Western Digital.

Related companies

$SNDK $MU