2026 Wafer Ramp Shows HBM Share of Memory Capex Hits 55%

2026 Wafer Ramp Shows HBM Share of Memory Capex Hits 55%

Track SK hynix's 1c wafer ramp to see HBM's grip on capex


SK hynix is scaling its 1c DRAM line from roughly 20,000 to as many as 190,000 monthly wafers by the end of 2026. That's an 8x to 9x jump on a single node. Meanwhile its quarterly operating margin has already climbed to near 80%, and DRAM volume alone doesn't explain that. So the real question is how much of that wafer number is actually an HBM capex program hiding behind a DRAM-node label, and how it stacks up against Samsung's narrower, qualification-gated 50% HBM capacity target once you put the two companies' numbers side by side.



Here's the part that changes how investors should read any capex headline out of Korea: HBM runs roughly 41% to 55% of total manufacturing cost across current HBM platforms, according to industry cost-structure analysis. That's cost share, not revenue share, and it reflects the stacking, bonding, and testing steps HBM needs that standard DRAM skips entirely. A DRAM die that ends up in an HBM stack carries several times the processing cost of the same die sold as a plain commodity module. So when SK hynix or Samsung put out a wafer capacity number, that figure masks a much bigger swing in where the money actually goes. A 1c wafer headed for HBM soaks up far more capital per unit than the same wafer headed for a legacy DDR5 module.



Samsung's approach looks different on paper. The company is targeting roughly 50% growth in HBM production capacity in 2026, a narrower, more surgical commitment compared to SK hynix's broad node ramp. Read it favorably and it's disciplined capital allocation: Samsung putting incremental capex specifically where margin and strategic positioning concentrate, instead of expanding DRAM capacity across the board and hoping HBM demand soaks up the extra output. Read it less favorably and the real constraint on Samsung's HBM business right now isn't installed wafer capacity at all. It's HBM4 qualification timelines. Capacity sitting idle ahead of qualification is capital spent with no revenue to show for it yet, and that gap is what investors need to track quarter to quarter before assuming Samsung's 50% target shows up in profit on the same schedule as SK hynix's wafer expansion.



Read Samsung's KRW 195.3 trillion quarter as an HBM mix signal


The wafer and cost-share numbers above explain why Samsung's and SK hynix's latest quarterly results look the way they do. Samsung Electronics' consolidated Q3 2026 revenue and operating profit reportedly grew substantially year-over-year, with figures circulating in some reports putting revenue around KRW 195.3 trillion and operating profit around KRW 107.4 trillion, though these figures have not been independently verified. These aren't organic-growth numbers. They're the signature of a company whose highest-margin product line, HBM, went from a niche allocation to the central profit driver inside a single fiscal year.



SK hynix's figures reportedly tell a parallel story with an even steeper curve. Some reports cite Q3 2026 revenue in the vicinity of KRW 104.6 trillion and operating profit around KRW 83.3 trillion, though again, unverified independently. Nine-month cumulative net income appears to have risen sharply compared with the prior-year period, which lines up with the margin expansion described above, though the precise figures aren't independently verified here. SK hynix's quarterly operating margin now sits close to 80%, and you don't get there on commodity DRAM pricing. The only real explanation for margins at that scale is that HBM, carrying its 41% to 55% cost share but commanding a far higher price premium, has become the dominant profit contributor, even if it isn't yet the dominant revenue line by unit volume. For investors, this margin level is the clearest confirmation available that HBM pricing power, not DRAM volume, now drives SK hynix's earnings.



Separate the wafer count from the HBM capex it hides


The margin numbers show HBM driving profit at both companies. The next question is how to read the capacity figures each one has published, since neither SK hynix's wafer count nor Samsung's capacity percentage states HBM exposure directly. If HBM cost share sits at the upper end of the 41% to 55% range for a given platform generation, and SK hynix is running an 8x to 9x wafer capacity expansion on the 1c node that feeds both HBM and legacy DRAM at the same time, a large chunk of that wafer expansion is, in effect, an HBM capex program wearing a DRAM-node label. The wafer count alone understates how concentrated the HBM exposure really is, because each HBM-bound wafer eats up more downstream capital in packaging and testing than a legacy-bound wafer off the same node.



Samsung's 50% HBM capacity target gives you a cleaner read precisely because it's stated in HBM-specific terms rather than node-wide wafer terms. But the qualification bottleneck means that capacity figure and revenue conversion don't sit on the same timeline. Comparing the two companies' capex efficiency means asking a different question for each one: for SK hynix, what share of the 1c wafer expansion is actually destined for HBM packaging versus legacy modules; for Samsung, when does HBM4 qualification turn installed capacity into shipped, revenue-generating product.



  • SK hynix 1c wafer ramp: ~20,000 to 160,000-190,000 monthly 300mm wafers by end-2026, an 8x-9x increase, split between HBM-bound and legacy-bound output.
  • Samsung HBM capacity target: roughly 50% growth in 2026, and the constraint is HBM4 qualification timing, not wafer availability.
  • HBM cost share: 41%-55% of total manufacturing cost across current HBM platforms, so capex dollars concentrate in HBM even when wafer counts look DRAM-generic.
  • SK hynix Q3 2026 operating margin: roughly 80%, which only makes sense with a mix shift heavily weighted toward HBM pricing power.

Anyone modeling these two companies' capex should stop treating published wafer or capacity figures as a direct stand-in for HBM investment. Discount a wafer number from SK hynix by the HBM-versus-legacy split within that node. Discount a capacity percentage from Samsung by qualification risk before assuming it converts into revenue.



Watch two signals before trusting either company's capacity target


Since neither company's headline figure holds up at face value, the practical task is figuring out what would actually make each one trustworthy. Investors should watch two separate signals going forward: SK hynix's disclosed HBM revenue mix within its DRAM segment, and the pace at which Samsung's HBM4 qualification milestones actually clear. That pace, not the 50% capacity target itself, determines when Samsung's capex allocation starts showing up in operating profit the way it already has for SK hynix. Until Samsung publishes a qualification-adjusted HBM revenue figure, treat the 50% target as a capacity ceiling, not a profit forecast. The original question was how much of SK hynix's wafer ramp is really HBM capex in disguise, and how it stacks up against Samsung's narrower target. The answer: SK hynix's near-80% operating margin shows HBM is already paying off inside its wafer numbers, while Samsung's 50% target remains unproven until qualification data, not capacity data, says otherwise. KOSPI investors pricing the two stocks on comparable multiples should weight SK hynix's margin trajectory more heavily than Samsung's stated capacity growth.


Related reading

Sources