SK Hynix's Premium Shrinks as Samsung's HBM4 Charge Begins

The number that matters

LS Securities knocked SK Hynix’s target price down by 27 percent — from 3.3 million won to 2.4 million won — on August 31. The headline cuts deep, but the real story sits two sentences below it. Analyst Jung Woo-sung didn’t downgrade the stock’s rating. He kept it at “buy.” What he changed was his assumption about how much extra profit SK Hynix can squeeze out of high-bandwidth memory.

The firm had been pricing in an operating profit margin of roughly 80 percent on next year’s HBM sales. After an industry check, it now expects that margin to hold closer to 60 percent. Same product category. Same customers. A quarter-point less profit per chip. Over billions in volume, that gap is where the valuation hit lives.

Why Samsung changes the math

SK Hynix has ridden a narrow lead into 2026, supplying the HBM stacks that power Nvidia’s latest AI accelerators. The company’s dominance was never questioned — it was priced into every forecast. Samsung, long the runner-up, has been quietly working through qualification hurdles with its HBM4 line.

Jung’s note signals that Samsung’s progress is no longer theoretical. The analyst wrote that “competition from rivals expanding HBM4 shipments and improving mass-production viability” raises the odds that major customers diversify their supply lines. That is the first concrete language from a Korean broker suggesting supplier diversification is moving from possibility to planning.

This is not a demand problem. Memory demand for AI remains insatiable. The problem is that SK Hynix can no longer rely on being the only game in town.

The margin floor that everyone was ignoring

The 80 percent margin assumption sounded plausible until you follow it to its logical endpoint. Jung pointed out the uncomfortable arithmetic: if HBM margins hit 80 percent, Nvidia would need to keep raising prices just to maintain its own 75 percent gross margin. Big Tech server budgets are already strained by memory cost inflation. Push HBM margins higher and you compress the very AI spending that drives demand.

Sixty percent, in Jung’s framing, is the new ideal — a level that keeps customers happy, preserves supplier returns, and leaves room for AI capex to continue. It is a sensible number. It is also a downgrade from the fantasy that SK Hynix could extract near-monopoly rents from a market that desperately needs competitive pressure to stay affordable.

The caveat matters: Jung did not rule out margin expansion again if yield improvements drive costs down. But the baseline assumption has shifted, and share prices adjust to baselines, not hope.

What this means for the supply chain

Samsung entering HBM4 volume seriously changes the architecture of the AI chip supply chain. Right now, the flow is simple: Nvidia designs the GPU, SK Hynix packages the HBM, and everyone takes their slice. Add a qualified second source and the model becomes tripartite — and each party recalculates.

For Nvidia, a second supplier is leverage. It weakens SK Hynix’s negotiating position on price and delivery terms. For SK Hynix, it means the premium collapses faster than demand growth can compensate. For Samsung, it means the memory division finally has a credible entry into the highest-margin segment of the AI hardware market — a segment that has been essentially off-limits until now.

The timeline is the critical variable. Samsung’s qualification process is not yet complete for all major customers. Jung’s language about “possible diversification” is deliberately cautious. But the direction is clear, and markets price direction long before the destination is certain.

The valuation framework, revised

LS Securities laid out a new PBR range for SK Hynix: 1.1 to 1.4 times through 2028 as the base case. If yields improve and AI demand spreads into new applications, the ceiling could reach 1.6 times. That is not a crash thesis — it is a normalization thesis. The company is still growing. The margin cushion is just thinner than the previous round of forecasts assumed.

What English-language desks have largely missed is the implication beyond SK Hynix’s balance sheet. When a single Korean broker revises an 80 percent HBM margin assumption to 60 percent and explicitly ties it to Samsung’s HBM4 progress, it is a signal that the competitive landscape is hardening in real time. Other analysts will follow. Other targets will adjust. The narrative that SK Hynix holds a permanent structural moat in HBM is losing its grip.

Who wins, who loses

Samsung wins most directly. Even a modest share of HBM4 volume at 60 percent margins is far more profitable than the near-zero it has been earning in this segment. Nvidia wins second — cheaper, more secure supply strengthens its own margin positioning against big-tech customers who are already pushing back on AI hardware costs.

SK Hynix does not lose in absolute terms. Its HBM shipments will keep growing. But the era of setting the price at the top of the market is over. The company that built its valuation on scarcity now has to compete on execution.

Big-tech buyers of AI infrastructure win third. Every point of margin compression on HBM is a point of cost relief on the servers they are already struggling to afford.

What happens next

Watch three things over the next quarter. First, whether Samsung announces formal HBM4 qualification with at least one major AI chip maker — that is the trigger that turns speculation into supply-chain reality. Second, whether LS’s 60 percent margin assumption holds or gets revised again as yield data arrives. Third, whether other Korean brokers issue similar adjustments. One firm’s estimate is a signal. Two or three is a trend.

The memory cycle has never been kind to complacency. SK Hynix dominated HBM for two years because the alternative did not exist. Now it exists. The premium that justified the old target price was always temporary. The question now is how quickly the market decides it was never as permanent as the consensus assumed.

The 27 percent cut is not the end of the story. It is the beginning of a new one.