Flash memory stores data as charge in a cell, and the industry's density roadmap has been a march of bits per cell: one, two, three, and now four. Each step trades endurance and write speed for capacity. QLC — quad-level cell — was long dismissed for data centers: too few write cycles, too slow on writes, too clever for its own good.
The AI era changed the calculation, because AI changed the workload.
Why it matters
Training and serving AI systems is a read-dominated existence: datasets are written once and read constantly, checkpoints are large and sequential, and retrieval corpora sit warm rather than hot. That is QLC's profile exactly. Meanwhile the power and space budgets of data centers are now the binding constraints, and a 100-plus-terabyte SSD replaces several hard drives in a fraction of the footprint and watts.
For storage buyers, the tiering math that held for a decade — hot data on TLC flash, everything else on disk — is being rewritten in real time.
How it works
Storing four bits means distinguishing sixteen charge levels per cell, which is why QLC writes are slow and cells wear out after far fewer cycles than triple-level flash. Vendors compensate with architecture: large SLC caches absorb write bursts, controllers spread wear across massive arrays, and the drives are positioned for workloads where a petabyte is written once and read a hundred times. Paired with the PCIe interface's bandwidth, a QLC drive delivers sequential reads no disk array can match at anything like the power draw.
The counter-move is equally interesting: hard drive makers, pushed out of the warm tier, are retreating up-capacity — 30TB-plus nearline drives using HAMR and similar recording technologies — defending the cold archive where cost per terabyte is the only metric that matters.
Evidence
Every major NAND maker now ships enterprise QLC lines, with 61TB drives common and 122TB-plus models in the channel; Solidigm's D5-P5336 and successors are the reference points of the category. Hyperscaler disclosures and vendor earnings describe QLC as the fastest-growing enterprise SSD segment, and storage analysts attribute it directly to AI data pipelines. Meanwhile HDD makers' own roadmaps show nearline capacity climbing even as unit shipments fall — the signature of a technology retreating to its last defensible tier.
The competing read
Flash optimists see an inevitable takeover: NAND cost per bit falls every generation, disk's fell off a cliff years ago, and the crossover spreads tier by tier. Disk defenders note the crossover point has been 'five years away' for fifteen years, that HDDs still hold the overwhelming majority of the world's exabytes, and that HAMR extends the cost lead for cold data another decade. Both are describing the same curve from different tiers — and the warm tier is where the growth is.
What happens next
Watch PLC — five bits per cell — in the labs, and watch the price curves: the day QLC's cost per terabyte crosses nearline disk's for mainstream capacities is the day the HDD market's last growth story ends. Also watch endurance engineering: if QLC write-endurance improves meaningfully, the warm tier's boundaries expand again.