SK hynix has reportedly fallen behind Samsung Electronics in qualifying hybrid bonding for High Bandwidth Memory (HBM), with the company's hybrid bonding yield remaining below the threshold for mass production, according to a TechPowerUp report.

The report frames Samsung as further along in deploying the wafer-level packaging method for upcoming HBM generations, where hybrid bonding has become a near-prerequisite for the highest layer-count stacks.

What is hybrid bonding and why it matters for HBM

Hybrid bonding connects stacked dies through direct copper-to-copper interfaces at the wafer level, eliminating the solder microbumps and underfill used in conventional thermo-compression bonding (TCB). Removing these intermediate layers reduces stack height, lowers thermal resistance, and increases interconnect density per square millimeter.

For HBM, hybrid bonding matters because DRAM stacks are growing taller with each generation. Current HBM3E parts ship in 8-Hi and 12-Hi configurations; the industry roadmap points to 16-Hi stacks for HBM4 and beyond. Conventional TCB runs into yield, warpage, and throughput problems as stack count rises, which is why hybrid bonding has emerged as the long-term stacking method for the tallest HBM parts.

Where the three HBM suppliers stand

The TechPowerUp report indicates that SK hynix's hybrid bonding yields remain short of mass-production requirements, while Samsung has achieved more mature internal results. Micron, the third major HBM supplier, has not been positioned in the report.

SK hynix currently holds the largest share of the merchant HBM market, having supplied the bulk of HBM3 and HBM3E used in AI accelerators from Nvidia and AMD. Samsung holds the second position in merchant HBM, with Micron in third. The reported difficulty with hybrid bonding does not affect SK hynix's existing HBM3E shipments.

The reported gap does shift the competitive calculation for HBM4, where qualification cycles are already active at major AI accelerator customers. Samsung has publicly discussed hybrid bonding as part of its memory packaging roadmap. SK hynix has not commented on the specific yield figures in the TechPowerUp report.

What changes if Samsung's lead holds

If Samsung maintains its hybrid bonding lead through HBM4 qualification, customers seeking the highest-stack-count parts could shift allocations or dual-source toward Samsung, particularly for AI accelerators designed around 16-Hi stacks.

SK hynix's response depends on how quickly it closes the yield gap, and on whether the first HBM4 wave requires hybrid bonding outright or can still rely on advanced TCB. Design wins for the first HBM4-based accelerators are being decided during the current qualification window, which makes the timing of any yield recovery commercially significant.

What to watch next

Three signals will determine whether the reported gap becomes a lasting shift or a temporary setback for SK hynix: published hybrid bonding defect-density figures from either Korean supplier, confirmation of whether the first HBM4 generation ships with hybrid bonding or TCB, and any change in HBM allocation at Nvidia, AMD, or custom accelerator customers during 2025.