Nvidia triple-qualifies HBM4 suppliers; SK Hynix, Samsung, Micron all production-ready for Vera Rubin Q3 ship
On June 5, 2026, Nvidia CEO Jensen Huang confirmed at Seoul's Gimpo Airport that all three major memory suppliers—Samsung Electronics, SK Hynix, and Micron Technology—have been qualified and are in active production for HBM4 memory chips destined for Vera Rubin AI servers. The confirmation ended months of supply-chain speculation and marked the first time Huang publicly acknowledged all three as cleared for simultaneous HBM4 delivery. Vera Rubin entered full production in June 2026 after Huang's GTC Taipei keynote on June 1, with first customer shipments scheduled for Q3 2026. Supply-chain analysts estimate SK Hynix holds 60–70% of allocated Vera Rubin HBM4 volume, Samsung 25–30%, and Micron the remainder.
HBM4 represents a structural break from HBM3E: it doubles the memory interface width from 1,024 bits to 2,048 bits and increases independent data channels from 16 to 32, delivering at least 2 terabytes per second of bandwidth per memory stack at JEDEC baseline. Each Vera Rubin GPU package uses 288GB of HBM4 in 8 stacks, delivering 22 TB/s memory bandwidth per package. A single Vera Rubin NVL72 rack contains 20.7TB of HBM4 capacity and delivers 1.6 PB/s aggregate HBM bandwidth—2.8x that of Grace Blackwell NVL72. Huang told reporters the second half of 2026 will be a major production ramp, with 2027 larger still.
For architects: this triple-qualification is the most decisive signal yet that HBM supply will be the dominant constraint through 2027. Huang's June 5 Seoul remarks—"All three vendors are in production, and they're all racing to support Vera Rubin"—signal Nvidia expects competition-driven volume growth from all three, but supply parity among vendors is impossible given SK Hynix's 60–70% allocation. Vera Rubin's Q3 2026 ship date is now locked; the challenge shifts to ODM ramp (Foxconn, Quanta, Wistron) and HBM4 yield and packaging speed. For infrastructure architects, Vera Rubin availability in H2 2026 will be the bottleneck for all AI data center expansion; chip availability is no longer the limiting factor—HBM packaging and integration speed are.
Sources
- Primary source
- finance.yahoo.com
“Jensen Huang certified Samsung, SK Hynix, Micron for HBM4 supply June 5; SK Hynix 60-70%, Samsung 25-30%, Micron remainder of Vera Rubin allocation”
- cnbc.com
“SK Hynix surged 12% after Micron earnings; Vera Rubin full production ramp with HBM4 supply locked, Huang said all three vendors in active production”
- vrlatech.com
“Vera Rubin NVL72: 72 Rubin GPU packages, 36 Vera CPUs, 260 TB/s aggregate NVLink, 3.6 EFLOPS NVFP4 inference, 20.7TB HBM4 capacity, 1.6 PB/s aggregate HBM bandwidth, ready H2 2026”