Bernstein analysts released a report on June 22 noting that conventional DRAM prices, after surging roughly 4.5x from Q3 2025 to Q2 2026, are likely to continue rising through 2027. The memory industry's 'super cycle' has not peaked, and conventional products are outperforming HBM in profitability.
Regular DRAM's Per-Bit Price Surpasses HBM
The report states that regular DRAM's average selling price per bit is now comparable to or higher than HBM. By 2026, regular DRAM's per-wafer revenue could be twice that of HBM, with significantly higher margins. At SK Hynix, HBM consumes about 23% of wafer capacity, while regular DRAM uses 77% yet generates higher profit per wafer.
HBM Needs Triple Price Hike to Catch Up
Bernstein calculates that HBM prices would need to triple to match regular DRAM's per-wafer revenue. This explains why while AI chip narratives focus on HBM, conventional memory remains the profit engine for major manufacturers. The analysts caution that pricing strategies won't be aggressive — simultaneous price hikes for HBM and regular DRAM could drive up AI server costs, raising cloud service fees and increasing tech giants' capex.
Taiwan Supply Chain Benefits, AI Cost Pressures Mount
The price surge has directly impacted Taiwan's memory supply chain — wafer fabs and packaging plants in Nanke are near full utilization. Nine major U.S. industries recently wrote to the Trump administration, warning that memory price inflation and shortages are hurting downstream manufacturing costs. SK Hynix and Samsung have accelerated their U.S. fab construction plans. The 4.5x rally from Q3 2025 to Q2 2026 has pushed the memory industry into a similar 'super cycle'. If prices continue rising in 2027, AI training and inference hardware costs could increase further, impacting pricing strategies for large models.