- SK Hynix is printing record profits from a supply shortage it is actively destroying by triple wafer capacity
- Q1 2026 operating profit hit ₩37.6 trillion ($24.5 billion), up 398% year-over-year - a number that exists precisely because HBM is constrained, not in spite of it
- Nvidia, which sources roughly 90% of its HBM from SK Hynix today, is expected to halve that share to around 50% by 2026 as Samsung and Micron catch up
- The company plans a $14 billion ADR offering this August, effectively using today's scarcity premium to finance capacity that erodes tomorrow's scarcity premium
- I rate SK Hynix a Hold at these levels: the HBM bull run is real, but the stock now prices perfection while the structural setup is rewiring itself
SK Hynix shares jumped roughly 15% last week on reports the company is tripling its wafer capacity for high-bandwidth memory - the specialized chip that powers every NVIDIA GPU in every AI data center. The market's reflex was instant: more capacity, more HBM, more profit. The stock has surged roughly 58% in the past month and cleared the $1 trillion market capitalization threshold.
I've been very surprised that investors are rewarding what is, in structural terms, a shareholder-value destruction machine disguised as a growth story. The false narrative here is that capacity expansion equals future profit. In memory semiconductors, it almost always means the opposite.
SK Hynix posted ₩37.6 trillion ($24.5 billion) in operating profit for the first quarter of 2026, a 398% year-over-year jump. The number is staggering. It is also almost entirely a function of HBM scarcity. High-bandwidth memory - stacked memory chips that feed AI processors at speeds conventional DRAM cannot match - is currently sold out through 2026. SK Hynix controls roughly 62% of the HBM market and supplies about 90% of NVIDIA's HBM demand. The operating profit explosion exists because demand vastly outstrips supply, not because SK Hynix has invented a perpetual profit engine.
That being the case, the logical investment question isn't whether SK Hynix is winning today. It's what happens when the company triples the capacity that is creating today's margins.
The answer is clear: the supply constraint disappears, prices normalize, and the profit margin that now exceeds 50% compresses toward historical averages. Memory semiconductors have an established cycle - prices spike when supply is tight, manufacturers overinvest, prices collapse when the new capacity comes online. SK Hynix is the beneficiary of the spike right now. The tripling of wafer capacity is the overinvestment phase.
Three structural forces make this worse than a standard cyclical play.
First, NVIDIA is actively diversifying its HBM supply chain. SK Hynix's share of NVIDIA's HBM purchases is expected to fall from roughly 90% today to around 50% by 2026. Samsung and Micron are both ramping production. SK Hynix will be tripling capacity into a market where its own customer is deliberately reducing reliance on it.
Second, Samsung - historically SK Hynix's rival and now its nearest competitor - is catching fast in HBM technology. Samsung's HBM market share has been declining, but the gap is narrowing. When two of the three memory giants are simultaneously expanding HBM capacity, the scarcity premium evaporates for everyone. Only SK Hynix and perhaps a couple of early HBM adopters may see some marginal gains from the current transition. The broader memory complex will face a capacity glut.
Third, the capital allocation is misaligned with shareholder returns. SK Hynix pays a fixed dividend of ₩1,500 per share plus 5% of annual free cash flow as a discretionary component. The trailing twelve-month dividend yield sits at roughly 2.84% against levered free cash flow of ₩25.8 trillion. That's a thin commitment. The company is reinvesting the lion's share of its record profits into capacity that will structurally depress the very margins generating those profits. Then, to fund even more expansion, it plans a $14 billion ADR listing on U.S. exchanges this August.
Compare that capital return philosophy with what matters to long-term income investors: companies that prioritize sustainable dividends and dividend growth represent a tangible commitment to shareholders. SK Hynix's dividend policy - a small fixed base with a discretionary kicker tied to free cash flow - reads as management reserving the right to spend every won on capacity while giving investors a rounding error.
The ADR offering complicates the picture further. Raising $14 billion by issuing new shares dilutes existing holders at the peak of the scarcity premium. The money raised will almost certainly flow into more fab expansion - reinforcing the cycle where today's profits from constrained supply fund tomorrow's oversupply. That is not a capital allocation strategy designed for shareholder returns. It is designed for market share.
I acknowledge the counterargument. AI infrastructure spending is still accelerating. NVIDIA alone is guiding for hundreds of billions in annual capex. If HBM demand grows fast enough, the tripling of capacity might not create a glut - it might just maintain equilibrium. The bull case assumes AI memory demand stays structurally tight through 2027 and beyond, absorbing all new supply without price erosion.
However, that assumption ignores what every memory investor should know from decades of industry behavior: capacity decisions are made in boom years, come online in two to three years, and land precisely when demand normalizes. The physics of memory manufacturing do not bend to investor optimism. When the 02H fab in Pyeongtaek and expansion projects come online in 2027 and 2028, the supply overhang will arrive regardless of AI demand growth projections made today.
Additionally, AI revenue cross-pollination works in both directions. The same hyperscaler capex driving SK Hynix's HBM demand is also funding custom silicon programs from Google, Amazon, and Microsoft - chips that may use less HBM per dollar of compute or different memory architectures altogether. The HBM story isn't one-directional, even if the current headline only points one way.
I rate SK Hynix a Hold at current levels. The HBM bull run is real, the company is executing well, and the Q1 2026 results validate the structural AI demand shift. But the stock has priced in perfection while the underlying dynamics are already rewiring: customers diversifying away, competitors catching up, capacity tripling into a market that is creating today's profits through scarcity. For investors who want exposure to the AI memory buildout with better capital alignment, the broader semiconductor ETFs or direct NVIDIA holdings offer cleaner structural positioning. SK Hynix is a company that is winning the present by funding the future that makes the present irrelevant.
That being the case, I would not buy at these levels, and I would not sell existing positions - but I would watch the NVIDIA supply share, the Samsung HBM ramp, and the 2027 capacity arrival dates with extreme caution. In memory semiconductors, the last company to invest wins the cycle and loses the next one. SK Hynix is currently the last company to invest. That is both the thesis and the trap.