Micron $MU ( ▼ 0.67% ) is basically making a long-term bet on one thing: the memory shortage isn’t going away anytime soon.
The company just signed a letter of intent to buy a fabrication plant from Powerchip Semiconductor Manufacturing Corporation in Taiwan for $1.8 billion, a move designed to expand DRAM output right when the AI supply crunch is expected to still be raging.
What Micron is actually doing here
Micron says this deal will help it “increase production and better serve customers” in a market where demand is still outrunning supply.
The key detail: management expects the purchase to begin adding to DRAM production in the second half of 2027.
So this is not a quick fix. This is Micron loading up capacity for the next phase of the AI infrastructure cycle.
Why this matters (this is a long memory supercycle)
Micron’s mid-December earnings and guidance lit a fire under memory and storage names because it confirmed what investors were starting to suspect:
AI demand is pulling memory forward faster than anyone modeled.
And it’s not just regular DRAM. The bigger issue is HBM, the high-bandwidth memory that basically acts like the “fuel line” feeding GPUs.
Demand is surging, supply is tight, and pricing power stays with the producers as long as the shortage persists.
JPMorgan’s math shows the shortage is massive
JPMorgan analysts led by Jay Kwon put hard numbers on the problem:
Even ignoring conventional DRAM supply-demand, they estimate the DRAM industry will be in a net deficit of about 180k wafer starts per month in 2027 just to resolve the HBM shortage.
That’s the part people keep missing. This is not a small gap.
They also highlighted how limited even big expansions really are:
Micron’s P5 fab at 50k wafer starts per month, even if it were fully dedicated to HBM, would only cover about 20% of the shortage by their calculation.
So even after Micron spends billions, the market can still stay structurally tight.
Bottom line
Micron isn’t buying a fab because it thinks demand will cool off.
It’s buying a fab because AI is turning memory into one of the most persistent bottlenecks in the entire chip stack.
And if supply stays constrained into 2027 and beyond, Micron gets exactly what it wants: sustained pricing power.