TrendForce released an extremely bullish take on memory

TrendForce says HBM prices have not fully caught up yet because the big suppliers usually negotiate HBM contracts annually. Meanwhile, conventional DRAM prices have been rising sharply since late 2025, making products like DDR5 64GB RDIMMs more profitable than HBM on a per-wafer basis in 1Q26
That creates a major incentive problem. If conventional DRAM becomes more profitable than HBM, suppliers may prefer to allocate more wafer capacity to conventional DRAM unless HBM prices rise significantly. TrendForce expects this to support much higher HBM contract prices in 2027
━━━━━━━━━━━━━━━
Main demand drivers
➡️ In 2026, HBM demand growth will mainly come from AI ASICs, where memory capacity per chip is moving from 96GB/192GB to 216GB/288GB
➡️ NVIDIA Rubin will also increase total HBM demand through higher shipment volumes, even if HBM capacity per GPU remains similar to the prior generation
In 2027, demand accelerates again because Rubin Ultra is expected to use 384GB of HBM per GPU, while Google TPUs and other AI ASICs continue scaling deployment volumes
━━━━━━━━━━━━━━━
Supply impact
TrendForce estimates that HBM wafer input will rise from:
⏩ 18% of total DRAM wafer input by end-2025
⏩ 22% by end-2026
⏩ 30% by end-2027
But because HBM uses larger die sizes and has lower bit output per wafer, HBM bit supply will only represent:
⏩ 8% of DRAM bit supply in 2025
⏩ 9% in 2026
⏩ 13% in 2027
This is the most important part. HBM takes a lot of wafer capacity, but it does not add much total DRAM bit supply. So the more HBM grows, the more it crowds out conventional DRAM capacity
━━━━━━━━━━━━━━━
This is extremely bullish for memory suppliers like $MU, SK Hynix, and Samsung
It supports the idea that memory is no longer just a cyclical commodity story. AI is creating a structural demand shock across HBM, RDIMMs, server LPDDR, and conventional DRAM