NAND flash cells physically die. Each program/erase (P/E) cycle rips electrons through a thin oxide layer, and after enough cycles that oxide leaks so badly the cell can't hold a charge. SLC endures ~100k cycles, MLC ~10k, TLC ~3k, QLC ~1k. If software wrote to the same logical block a million times, that block would be dust within a day. Yet your SSD survives years. The trick is the Flash Translation Layer (FTL), a firmware layer that lies to the operating system about where data physically lives.

When the OS writes to LBA (logical block address) 42, the FTL doesn't overwrite the previous physical location — it writes to a fresh, unused page and updates an internal map: LBA 42 → physical page 8,193,447. The old page is marked stale. This is called out-of-place writes, and it's mandatory because NAND can only be programmed page-at-a-time (typically 16KB) but erased block-at-a-time (typically 4MB of many pages). You can't overwrite a page — you must erase the whole containing block first.

Wear-leveling comes in two flavors:

• Dynamic: Only shuffles hot data across free blocks. Cold data (your OS install from 2019) sits on the same physical block forever, so those blocks age slowly while hot ones age fast.
• Static: Periodically moves cold data onto worn blocks to free the young blocks for hot writes. This levels wear across the whole drive at the cost of extra writes.

Garbage collection is the reaper: when free blocks run low, the FTL picks a block with many stale pages, copies the still-valid pages elsewhere, then erases the block. This causes write amplification (WA) — the ratio of physical NAND writes to host writes. WA of 3 means every 1 GB the OS writes causes 3 GB of NAND wear.

Rule of thumb — drive lifetime in TB written (TBW):

TBW ≈ (Capacity × P/E cycles) / Write Amplification

A 1 TB TLC drive with 3000 P/E cycles and WA of 3: TBW ≈ (1 × 3000) / 3 = 1000 TB written. At 50 GB/day that's ~55 years. At the sustained 500 MB/s of a database log, ~23 days.

Real example: Samsung's 980 PRO 1TB is rated for 600 TBW. Datacenter Optane drives hit 60+ PBW because 3D XPoint has ~10⁷ endurance. This is why enterprise drives cost 5× consumer drives with identical capacity — they over-provision 20-30% of NAND as spare blocks, reducing WA dramatically because the FTL always has clean blocks to write into.