TrendForce just lifted 2027 HBM price targets again. Blended ASP now up 121% year-over-year.
The latest note from September 29: AI servers are pulling HBM and conventional DRAM into a brutal fight for the same advanced wafers. Next-gen yields take time to ramp. Supply stays tight straight through 2027. Suppliers are holding firm on 2027 HBM price hikes. The driver? Shortage plus a richer HBM4 mix and HBM4e ramping in the second half of 2027.
GPU and ASIC makers are now looking at fewer HBM layers per device. Not because demand collapsed. Because bits are scarce and system cost is exploding.
An 8-high stack uses fewer DRAM dies than a 12-high, cuts cost per stack, and lets the same wafer output feed more GPUs. TrendForce expects 8-high to become a 2027 priority.
But here's the twist: that does not cheapen the gigabit. Every stack still needs one base die. An 8-high spreads that fixed cost over fewer DRAM layers. Per-Gb pricing is expected 10–20% above 12-high in 2027.
So de-spec can ease GPU BOM pressure. It does not reverse the HBM price wave. Fewer layers, higher price per bit, still not enough wafers.
If 8-high is 10–20% more expensive per Gb, does 2027 revenue rise because stacks get shorter or because every remaining bit just costs more?
$MU $SKHY $DRAM
The latest note from September 29: AI servers are pulling HBM and conventional DRAM into a brutal fight for the same advanced wafers. Next-gen yields take time to ramp. Supply stays tight straight through 2027. Suppliers are holding firm on 2027 HBM price hikes. The driver? Shortage plus a richer HBM4 mix and HBM4e ramping in the second half of 2027.
GPU and ASIC makers are now looking at fewer HBM layers per device. Not because demand collapsed. Because bits are scarce and system cost is exploding.
An 8-high stack uses fewer DRAM dies than a 12-high, cuts cost per stack, and lets the same wafer output feed more GPUs. TrendForce expects 8-high to become a 2027 priority.
But here's the twist: that does not cheapen the gigabit. Every stack still needs one base die. An 8-high spreads that fixed cost over fewer DRAM layers. Per-Gb pricing is expected 10–20% above 12-high in 2027.
So de-spec can ease GPU BOM pressure. It does not reverse the HBM price wave. Fewer layers, higher price per bit, still not enough wafers.
If 8-high is 10–20% more expensive per Gb, does 2027 revenue rise because stacks get shorter or because every remaining bit just costs more?
$MU $SKHY $DRAM
