In memory semis, the Big Three are tilting capex heavily toward DRAM. Why are they pouring money into DRAM while staying 'lukewarm' on NAND?
On the demand side, AI has lit a fire under DRAM. HBM is indispensable to AI accelerators, and DDR demand has spiked with AI inference servers, pushing supply/demand into deficit and prices sharply higher.DDR5, for example, is up more than 5x, driving outsized margins.
NAND demand is rising mainly as AI servers displace HDDs and via additions tied to the RubiN architecture. The growth is real, but urgency and pricing power lag DRAM.
On the supply side, DRAM is hard to add: new lines typically require heavy capex and long lead times. Early movers can grab share and build moats.By contrast, NAND is easier to scale. Even without greenfield fabs, tool upgrades and higher layer counts lift output, enabling a quick ramp even if capex is deferred.
Competition in NAND is more fragmented, with more players beyond the Big Three, so extra capex does not necessarily translate into outsized profits. DRAM, where the Big Three control over 90% of the market, is more oligopolistic and offers steadier returns.Net-net, they rationally skew capex toward DRAM. They 'overweight' DRAM and 'underweight' NAND.
In short, DRAM offers stronger earnings leverage and more urgent demand. NAND is easier to scale and faces tougher competition.
