Industry Sources Raise HBM4 Concerns Over NVIDIA Vera Rubin Ramp
Suppliers of 800-volt power systems, M9 server-board materials and optical components could face changing demand expectations if the production ramp shifts.
Unnamed industry sources have raised concerns that a high-bandwidth memory (HBM4) bottleneck for NVIDIA’s Vera Rubin platform could push mass production into the first half of 2027, beyond its previous second-half 2026 shipment outlook.
The sources described GB300 servers as NVIDIA’s main shipment focus while the company waits for Vera Rubin production to ramp. A shift to early 2027 would put mass production later than the company’s previously outlined shipment schedule.
The reported concern contrasts with earlier statements from NVIDIA founder Jensen Huang. In June, Huang said HBM4 from SK Hynix, Samsung Electronics and Micron had passed qualification and entered production. In July, he rejected reports of a Rubin delay and said the platform had entered mass production.
The Vera Rubin NVL72 is already available to cloud customers through CoreWeave. The reported supply concern centers on the mass-production schedule.
The possible effects across Taiwan’s supply chain vary. Unnamed analysts assess that Hon Hai and Quanta may be less exposed because they have completed production learning curves and can deliver at scale. Suppliers of 800-volt DC power systems, M9 server-board materials and optical components tied to co-packaged optics (CPO) could face revised demand expectations and capacity plans if the ramp shifts.
Those components are tied to platform upgrades as computing performance and power needs rise. The reported supply-chain assessment identifies power systems, board materials and optical components as areas where expectations were linked to Vera Rubin’s production ramp.
NVIDIA is scheduled to release financial results and hold a conference call Tuesday, Nov. 17. The production timeline could be addressed during the event.