SK hynix has reportedly started shipping samples of its cutting-edge 12-high HBM4E memory, a significant move that transforms the AI memory race into a direct head-to-head contest with Samsung.
This isn't just about catching up. Samsung claimed the 'industry-first' title by announcing its HBM4E samples on May 29. However, SK hynix's entry just a few weeks later shifts the entire narrative. The competition is no longer about press release dates; it's now a technical battle over who can deliver the most reliable, high-performance memory that passes the rigorous qualification process for NVIDIA’s next-generation Vera Rubin AI platform.
Several key events led to this moment. First, Samsung’s announcement created immense pressure on SK hynix to accelerate its own timeline, which originally guided for samples in the second half of 2026. Second, NVIDIA CEO Jensen Huang recently confirmed that Samsung, SK hynix, and Micron are all certified for the base HBM4 memory. This leveled the playing field and put the spotlight squarely on the more advanced HBM4E.
Furthermore, the technological hurdles are immense. Advanced packaging technology like TSMC's CoWoS is essential, and managing the heat and stability of tall 12-layer memory stacks is a major engineering challenge. This means the real winner will be determined by performance in customer systems, not just by shipping a sample.
The stakes couldn't be higher. The demand for HBM is so intense that customers are reportedly offering to pre-pay for manufacturing equipment just to secure supply. With each NVIDIA Rubin GPU potentially containing thousands of dollars worth of HBM, winning the qualification race means securing billions in revenue through long-term supply agreements. SK hynix's move ensures it remains a top contender in this high-stakes game.
- HBM (High Bandwidth Memory): A type of high-performance computer memory used alongside GPUs to process massive amounts of data quickly, essential for AI applications.
- Qualification: The process where a customer (like NVIDIA) rigorously tests a supplier's component (like HBM) to ensure it meets all performance, reliability, and compatibility standards before approving it for mass production.
- CoWoS (Chip-on-Wafer-on-Substrate): An advanced packaging technology used to integrate multiple chips, like a GPU and HBM stacks, closely together on a single substrate to achieve higher performance.