
According to South Korean media, Micron's HBM4 products are struggling to meet NVIDIA's stringent performance and power-efficiency requirements, potentially forcing the company to redesign its HBM4 chip architecture. If true, this could delay Micron's mass production by up to nine months, pushing its HBM4 rollout into 2026 and causing it to miss NVIDIA's orders.
In contrast, SK hynix announced in September that its next-generation HBM4 has completed development and entered mass production, fully meeting client performance demands and supporting industry-leading speeds. Shipments are scheduled to begin in Q4 2025, with plans to expand sales significantly next year. Reports indicate that SK hynix's HBM4 has already passed NVIDIA's validation.
Samsung, aiming to maintain its leadership in DRAM, plans to leverage its advanced 1c DRAM-based HBM4 for large-scale production, ensuring a competitive edge over SK hynix. Industry sources suggest Samsung's 1c DRAM yield has exceeded 50%, while its HBM4 logic chip yield has reached an impressive 90%. Samsung's HBM4 is expected to launch in Q4 2025, with no current signs of delays.
Experts note that HBM4 is a critical component for AI and data center high-power computing, demanding significantly higher bandwidth, power efficiency, and stacked integration compared to previous generations. SK hynix was the first to complete global mass production of HBM4 last month, while Samsung showcased a working HBM4 at the 2025 Semiconductor Expo (SEDEX) this month and has already begun mass production preparations. The technological gap between the two companies is widening.
Analysts warn that if Micron continues to lag behind SK hynix and Samsung, it could miss not only key NVIDIA HBM4 procurement opportunities but also upcoming AI data center orders. Market expectations suggest that SK hynix and Samsung will continue to dominate the HBM4 market in the near term, and whether Micron can re-enter the competition in 2026 will depend on its progress in redesigning HBM4 and improving yields.