Galaxy Securities predicts that the surge in AI computing power construction will lead to a prolonged increase in memory prices, with High Bandwidth Memory (HBM) production share expected to rise significantly. NVIDIA's upcoming Rubin Ultra platform will increase HBM capacity per GPU to 384GB, further boosting demand alongside growing AI ASIC shipments. TrendForce forecasts HBM's share of total DRAM production to grow from 18% in 2025 to approximately 30% by 2027, with its bit supply share increasing from 8% to about 13%. The report also highlights 3D stacked DRAM as a key technology for overcoming traditional memory bottlenecks driven by the growth of edge AI. AI
IMPACT Increased demand for HBM and 3D stacked DRAM will likely drive innovation and potentially higher costs in AI hardware components.
RANK_REASON The item is an analysis from Galaxy Securities about market trends in AI hardware, not a direct release from a frontier lab or a major industry event.
- AI ASIC
- dynamic random-access memory
- Galaxy Securities
- graphics processing unit
- High Bandwidth Memory
- NVIDIA
- Rubin ultra
- TrendForce
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →