The hardware behind the future


Loose Change: Analysts Now Have a Date for the End of the DRAM Squeeze

UBS pegs supply-demand balance at the second quarter of 2028, and the wafer math behind that date explains why memory stocks got cheaper while the shortage got longer.

Loose Change: Analysts Now Have a Date for the End of the DRAM Squeeze

The firms that count memory wafers for a living have converged on an answer to the question every infrastructure buyer is asking, and it is not soon: UBS does not expect DRAM supply and demand to balance until the second quarter of 2028, a call that lands while Micron tells investors that tight conditions should persist beyond calendar 2027.

The arithmetic behind that date is plain. TrendForce and Counterpoint Research estimates cited by 24/7 Wall St. have HBM consuming 30 percent of total DRAM wafer capacity by the end of 2027, with supply covering only 60 percent of projected demand by then. Every wafer that becomes HBM is a wafer that stops being the commodity DRAM that servers and phones buy in volume, and the same research projects memory absorbing 40 percent of AI capital spending by 2030.

New capacity cannot arrive faster than the concrete cures. A fab needs 12 to 24 months to reach peak output, so money committed now produces bits in late 2027 at the earliest. Industry capital spending is rising from 53.7 billion dollars in 2025 to 61.3 billion in 2026, and a 14 percent increase is restraint by the standards of past booms. Bank of America goes further, estimating SK Hynix may deliver one sixth of its originally planned capacity additions by 2028.

Translated, the suppliers learned their lesson. Samsung holds roughly 38 percent of the DRAM market, SK Hynix 29 percent, Micron 22 percent, and none of the three has an incentive to build its way out of the best pricing environment in years. The last aggressive spend into a shortage manufactured the 2023 glut, and institutional memory of those write-downs is doing as much to keep prices high as any AI order book.

The tape, meanwhile, argues with the calendar. Memory stocks trade at six to seven times forward earnings after a pullback that took the Roundhill Memory ETF from a 191 percent gain to 91 percent for the year, while Micron's own shares remain up 176 percent. A low multiple on cyclical peak earnings is the oldest trap in this sector, since the ratio flatters exactly when earnings sit at highs. If the UBS date holds, though, the tightness outlasts every 2027 budget cycle now being written.

Bootnote: the purest expression of this trade is an exchange-traded fund whose ticker is simply DRAM. It holds the memory and storage complex and nothing else, which makes its 91 percent year a live gauge of how much shortage the market still believes in. When a commodity gets its own ticker, the cycle has usually been generous for a while.

Leave a Reply