Flash stopped being a rounding error: Gartner sees NAND per-gigabyte prices up over 200%

Gartner projects NAND average selling prices per gigabyte to rise by more than 200 percent this year, directly on top of the 40 to 80 percent surges recorded through late 2025, according to an opinion column published by DataCenterDynamics ^{[1] [2]}. That breaks the assumption most storage plans were built on, which is that flash gets cheaper on a predictable curve, and it converts capacity planning from a capex line into a supply exposure with lead times of up to half a year ^{[5] [6]}.

The compounding is the part worth putting in a spreadsheet. Applying the 2025 range to the projected 2026 spike puts cumulative per-gigabyte pricing at roughly 4.2 to 5.4 times a pre-surge baseline, an increase of about 320 to 440 percent ^[1]. Gartner expects the dynamic to endure through 2027 before softening across 2028 and 2029 ^[3], which means any plan that treats this as a one-quarter squeeze is planning for relief no earlier than 2028 ^[2]. The column also attributes part of the volatility to geopolitical disruption of critical energy sources feeding manufacturing and supply chains ^[4], which is the kind of input that does not respond to a purchase order.

The operational consequence is scheduling, not sentiment. If delivery can take six months, capacity you need in the second half of the year has to be committed in the first half, two quarters ahead of the requirement ^[3]. Contracts that renew on a calendar cycle will therefore be negotiated after the decision that mattered has already been made or missed.

The remedy the column argues for is tiering: unless workloads genuinely require the lowest access latencies, combine flash, HDD, and tape in a single unified system rather than defaulting to all-flash ^{[7] [8]}. Its performance claim is that a scale-out design gets throughput from parallelism across nodes, surgical use of flash for metadata, and an efficient software data path, rather than from an expensive homogeneous flash tier ^[9]. Flash is reserved for hot objects, metadata, active AI pipelines, and indexing; HDD carries bulk capacity; tape holds long-term retention at the lowest cost per gigabyte ^[10]. Separately, it argues for software-defined, hardware-agnostic buying so a team can pivot from one vendor's SKU to another without rewriting the storage stack ^[11].

Treat that as the argument it is. The piece provides no cost-per-gigabyte comparison for HDD or tape, and no estimate of migration or retooling cost ^[12], which are exactly the numbers that decide whether re-tiering beats simply buying flash early at a bad price. The directional logic holds regardless: buying premium media for cold data is now a much more expensive habit than it was eighteen months ago ^{[7] [10]}.

Three things to track. Whether realised NAND pricing lands near the projected 200 percent or overshoots ^[1]. Whether quoted lead times stretch past six months, which is the earliest signal that a committed plan slips ^[5]. And whether the 2028-2029 softening Gartner forecasts starts appearing in supplier guidance, or gets pushed out again ^[3].