The cheapest memory chips are suddenly the hardest to buy
NOR flash and SLC NAND contract prices more than doubled in the first half of 2026, and analysts expect another 120-170% rise as AI demand soaks up the wafers.
The memory crunch has been framed as a graphics card problem, a phone problem and a data centre problem. The chips that will sting most are the ones nobody puts on a spec sheet: NOR flash and SLC NAND, the cheap, unfashionable memory that holds boot code in cars, routers and industrial hardware.
What the numbers say
TrendForce found that NOR flash and SLC NAND contract prices rose by more than 100 percent in the first half of 2026, and it expects SLC NAND to climb another 120 to 170 percent in the second half against the first. BNP Paribas forecasts the average NAND price reaching $279.50 per terabyte in 2026, up from $73.10 in 2025, and JPMorgan expects the shortage to run at least two more years, with customers receiving 70 to 80 percent of the orders they place.
Why nobody wants to build more of it
Jim Handy of Objective Analysis told Tom's Hardware that the entire SLC NAND market is probably worth under a billion dollars a year, which is small enough that nobody wants to fund new capacity. A 12-inch wafer used for mainstream NAND can generate close to $20,000 in revenue, according to TrendForce's Bryan Ao; the same wafer turned over to SLC brings in $6,000 to $8,000. Micron, Kioxia and SK hynix have been shifting capacity accordingly, and lead times for some manufacturing equipment have stretched to between 12 and 15 months.
That leaves an awkward trap: too little demand to justify a new factory, but more demand than the shrinking supply can serve. Handy's summary is the one to remember, because chipmakers either accept thinner margins or hand the bill to you. He compares the AI buildout to the late-1990s internet infrastructure boom, with the same risk of overshooting what the market can actually support.
Our opinion
Every memory shortage arrives with a comforting story attached: capacity will show up, prices will normalise, buy later. This one breaks the pattern, because the chips in trouble are too cheap for anyone to want to expand capacity for them and too embedded to design out quickly. Kioxia pitching serial SLC NAND as a NOR substitute is a genuine option, and also a reminder that swapping a component means qualification work rather than a supply chain favour. If you specify automotive, industrial or networking hardware, treat memory as a design risk now rather than a line to renegotiate in six months.