A single comment from Nvidia CEO Jensen Huang has sent ripples through the global semiconductor market.
His confirmation that Samsung, SK hynix, and Micron are all qualified and producing HBM (High Bandwidth Memory) for Nvidia’s next-generation Vera Rubin AI platform is a game-changer. Why? Because it directly addresses the biggest question surrounding Rubin's production ramp-up. Nvidia recently announced that Rubin is in "full production," but building these powerful AI systems requires a massive, stable supply of HBM.
For a while, the primary bottleneck in AI chip production was advanced packaging, known as CoWoS. However, with TSMC significantly increasing its capacity and improving yields to over 98%, that constraint has eased. This shifted the spotlight squarely onto HBM: could suppliers produce enough of it, fast enough, to meet Nvidia's aggressive timeline?
Tracing back the events shows a clear causal chain. First, Samsung had previously struggled to pass Nvidia's HBM qualification tests, creating skepticism about its role. Second, after persistent development, Samsung successfully qualified its HBM3E chips in late 2025, paving the way for its return. Third, during this time, SK hynix cemented its leadership with HBM4 technology, while Micron also emerged as a viable competitor. With Nvidia's Rubin production announcement, the demand for HBM became too large for any single supplier to handle.
Huang's statement was the final piece of the puzzle. By confirming that "all three" vendors are shipping, he signaled that Nvidia has successfully de-risked its supply chain. It no longer depends on a single source. For the memory makers, however, this marks the beginning of intense competition over market share and pricing. This is precisely why Samsung and SK hynix's stock prices fell on the news—investors are pricing in the new competitive reality.
In essence, Huang's confirmation has shifted the narrative from supply chain risk to supply chain competition. The focus is no longer on whether Nvidia can get enough HBM, but on which of the three giants will win the largest share of the lucrative HBM market.
- Glossary
- HBM (High Bandwidth Memory): A high-performance memory standard that involves stacking multiple memory chips vertically to achieve much higher data transfer speeds, crucial for AI accelerators.
- CoWoS (Chip on Wafer on Substrate): An advanced packaging technology used to integrate multiple chips, like GPUs and HBM, onto a single substrate, enabling high performance.
- ASP (Average Selling Price): The average price at which a particular product or commodity is sold.