According Industry sources Sunday, Samsung Electronics recently received the strongest evaluation among memory makers in Nvidia’s HBM4 System-in-Package (SiP) testing, delivering superior operating speed and power efficiency.
Nvidia’s requested HBM4 volumes from Samsung for next year are also said to have exceeded the company’s internal expectations, fueling optimism for a strong comeback for the chipmaker amid dire memory shortage from legacy to premium hyperscale chips.
The developments narrow the lead held by SK hynix, the pioneer and market leader of HBM, as competition intensifies to supply Nvidia’s next-generation Rubin AI accelerator.
Nvidia plans to launch Rubin in the second half of next year.
Given that HBM shipments typically begin six to seven months before an AI accelerator launch, full-scale deliveries could start as early as the second quarter, placing the current negotiation window at a critical juncture.
As the race accelerates, both Samsung and SK hynix have begun delivering paid final HBM4 samples to Nvidia, moving beyond free prototypes and into a commercially driven, pre-contract phase. Industry officials say paid samples typically indicate that performance has largely met customer specifications and that pricing and volume talks are under way.
Market tracker TrendForce said the final outcome is likely to become clearer by mid-to-late quarter, once quality qualification is completed and supply contracts are formally finalized. Volumes and pricing are widely expected to be locked in during the first quarter of 2026.
SK hynix confirmed during its third-quarter earnings call that its HBM4 mass-production system is already in place. Industry sources have dismissed rumors of major redesigns, saying the company is now making only minor performance tweaks.
Shipments are expected to begin in the fourth quarter, with a full sales ramp next year. Local media reports say SK hynix is currently supplying around 20,000 to 30,000 final HBM4 samples to Nvidia for validation.
Samsung, meanwhile, appears to be rebounding after losing ground in the HBM3E cycle. Industry sources say its HBM4 supply talks with Nvidia are nearing completion, positioning Samsung as the likely second-largest supplier after SK hynix. Paid HBM4 samples have already been delivered and are undergoing quality checks within Nvidia’s Rubin platform.
Samsung’s Device Solutions division has also cleared Production Readiness Approval (PRA), the final internal gate before volume manufacturing. HBM4 is expected to deliver roughly 60 percent higher bandwidth than current HBM3E products, with Samsung’s early samples already exceeding Nvidia’s next-generation requirement of 11 gigabits per second per pin.
The company is using improved 1c-class DRAM combined with a 4-nanometer logic base die, a configuration that helps close the performance gap with rivals while keeping power consumption and thermal output in check at higher speeds. Once Nvidia gives final approval, Samsung can move straight into mass production, with manufacturing lines already prepared.
Samsung had already signaled progress during its third-quarter earnings call on Oct. 30, saying HBM4 samples were being shipped to major global customers alongside mass production of HBM3E. The company also confirmed that its foundry arm will prioritize stable 2-nanometer gate-all-around output and HBM4 base-die production in 2026, alongside the ramp-up of its new Taylor, Texas fab.
KB Securities in recent report projected that Samsung and SK hynix will command more than 90 percent of global HBM4 demand, with Micron Technology’s share in Nvidia’s Rubin platform likely to remain below 10 percent.
TechInsights estimates that Samsung and SK hynix will account for about 80 percent of the HBM market by 2027, when HBM4 becomes the mainstream product. Industry sources say Micron’s initial HBM4 supply to Rubin is expected to be limited, partly due to performance constraints linked to its logic-die manufacturing approach.
Copyright ⓒ Aju Press All rights reserved.