HBM4 to drive over 30% price premium amid rising complexity and AI demand

According to TrendForce, the demand for AI servers is pushing the development of HBM4, with major suppliers advancing their product roadmaps. HBM4 features a more complex chip design with a larger die size due to an increased I/O count, raising production costs. Some vendors are shifting to a logic-based base die architecture to improve performance, further contributing to higher costs. Compared to HBM3e’s 20% price premium, HBM4 is expected to exceed 30% due to its manufacturing complexity.
NVIDIA’s Rubin GPU and AMD’s MI400 series, both slated to incorporate HBM4, were announced as upcoming AI chip solutions. HBM4 doubles the I/O count from 1,024 to 2,048 compared to prior generations, maintaining a data transfer rate above 8.0 Gbps, matching HBM3e. This allows HBM4 to achieve twice the data throughput at the same speed due to its higher channel count.
Unlike HBM3e’s memory-only base die, which serves as a signal pass-through, SK hynix and Samsung are working with foundries to implement a logic-based base die for HBM4. This design enables tighter integration with SoCs, reducing latency, improving data path efficiency, and enhancing stability for high-speed transmission.
TrendForce predicts HBM shipments will exceed 30 billion gigabits by 2026, with HBM4’s market share growing steadily and surpassing HBM3e as the dominant solution by t...