Nvidia CEO Jensen Huang said Wednesday that Samsung, SK Hynix, and Micron are qualified to supply HBM4 memory for the Vera Rubin AI platform [1, 2, 3].
This certification secures the critical high-bandwidth memory supply chain needed to power Nvidia's next generation of AI accelerators. Without these specialized components, the company cannot meet the escalating computational demands of large-scale artificial intelligence models.
Speaking on Bloomberg Television on June 3, 2026, Huang said that the "big three" memory chipmakers have been cleared for the HBM4 supply [1, 2]. The move ensures that the Vera Rubin platform has the necessary hardware to maintain its performance trajectory as AI workloads grow more complex [3, 5].
Deliveries for the Vera Rubin AI platform are expected to begin in the third quarter of 2026 [4]. This timeline puts pressure on the approved suppliers to scale production rapidly to meet the anticipated launch window [4].
To support this demand, SK Hynix has pledged to double its current wafer capacity for HBM4 [6]. This expansion reflects the structural shift in memory demand, where high-performance AI chips require significantly more bandwidth than traditional computing hardware [6].
While some reports have suggested a conflict regarding Micron's status, other industry sources confirm the company is among the three cleared suppliers [2, 7]. The inclusion of all three major players reduces the risk of supply chain bottlenecks that could delay the rollout of the Rubin architecture [2, 3].
Nvidia continues to diversify its supplier base to avoid over-reliance on a single vendor. By qualifying Samsung, SK Hynix, and Micron simultaneously, the company creates a competitive environment that may drive down costs and accelerate technical iterations of HBM4 memory [3, 5].
“Nvidia cleared Samsung, SK Hynix, and Micron as the three biggest memory chipmakers to supply HBM4”
By qualifying the three largest memory producers, Nvidia is mitigating the risk of a single-point failure in its supply chain for the Vera Rubin platform. The pledge by SK Hynix to double its capacity underscores the massive scale of memory required for next-generation AI, suggesting that HBM4 will be the primary bottleneck for AI hardware scaling in late 2026.
