Dell CEO warns memory shortage will worsen as $95 billion AI server order backlog awaits DRAM and HBM - TweakTown

The $95B backlog is a liquidity trap, not a growth story. The structural issue: Samsung, SK Hynix, and Micron are redirecting 70%+ of advanced wafer capacity from DDR5 toward HBM3E/4. This is a permanent reallocation, not a cyclical dip. Dell sits in the second tier of HBM allocation—hyperscalers have locked direct supply contracts, leaving OEMs with residual, volatile inventory. Technically, HBM4's TSV stacking jumps from 8 to 12-16 layers, pushing yield ramp beyond 18 months. Layered on top: CoWoS-L advanced packaging remains a concurrent bottleneck. The 2025H2–2026Q1 supply gap will not self-correct. Competitive divergence hinges on upstream binding depth. Lenovo's proximity to Taiwan, China's local packaging ecosystem grants delivery flexibility Dell's North American assembly model lacks. Supermicro and HPE face identical constraints without differentiated hedges. 12-24 month outlook: Until Micron's new fabs come online in 2026, HBM premiums will hold at 40-60% above spot. If Dell's backlog slips beyond two quarters, customer penalty clauses and revenue recognition risk trigger simultaneously. The real inflection: AI infrastructure economics is shifting from compute race to memory-bottleneck economics. That reframe changes capital allocation across the entire stack.