SK hynix Is Sold Out Through 2029 — and Buying Back Its Own Stock

An 80% operating margin, KRW 98.7 trillion of net cash, and a management team that calls the shares undervalued, while the AI hardware tape quietly de-rates

000660.KS · Earnings Call · 2026-10-27

A record quarter that the market refused to pay for

SK hynix did not just beat expectations in the third quarter of 2026 — it broke the shape of its own income statement. Revenue rose 32% sequentially and 328% year-on-year to KRW 104.6 trillion, the first KRW 100 trillion quarter in company history. Operating profit reached KRW 83.3 trillion at an 80% operating margin, with EBITDA of KRW 87.7 trillion (84% margin). HBM4 entered full mass production in the quarter, server DDR5 shipments rose more than 40% sequentially, and DRAM ASP climbed by a high-20% percentage rate for a second straight quarter. “In the third quarter, tight supply and demand conditions continued due to growing demand for AI memory, leading to sustained rise in memory prices.” — Woo-Hyun Kim, CFO · 2026-10-27 And yet the number that matters most for the equity story sits below the operating line. Non-operating expenses totalled KRW 32.6 trillion — KRW 11.6 trillion of FX-related losses from a stronger won, and KRW 22.9 trillion of valuation losses on investment assets, overwhelmingly the Kioxia stake. Net income was still KRW 39.8 trillion, a 38% net margin. Read that again: a third of the quarter's operating profit was consumed by mark-to-market damage on an equity holding, and the company still earned more in three months than most semiconductor firms earn in a cycle. The Kioxia stake cuts both ways, and management knows it. Realised gains on that investment are now being handed back as a special dividend, while the unrealised mark is a live grenade in the P&L. The shareholder return package is enormous: approximately KRW 100 trillion for fiscal 2026, split between KRW 30 trillion or more of cash dividends and KRW 60 trillion or more of share acquisition and cancellation, including the KRW 40 trillion buyback completed last week. The third-quarter dividend alone is KRW 20,000 per share — half ordinary, half special from Kioxia. That is why share acquisition now ranks as the second-most-prominent theme on an SK hynix earnings call, and why cash dividend has forced its way into the top ten. A memory manufacturer is now speaking the language of a capital-return vehicle.

The supply-side argument management wants you to accept as structural

Underneath the returns arithmetic is a simpler claim: this cycle will not end the way the last four did, because the demand is contracted and the supply is throttled by physics and by choice. Management said it has signed long-term agreements with 14 key customers, including major global CSPs, typically on five-year terms, with deposits to enforce purchase plans. More than half of 2027 and 2028 capacity is already committed. The CEO's summary of the position was unusually blunt:

The volume already secured under LTAs and the demand currently being discussed with customers taken together means that our DRAM and HBM volumes for the next 3 years through 2029 are effectively sold out.
Unknown Executive, Executive · 2026-10-27

That is demand visibility of a kind this industry has essentially never had. It is also the reason the word long term demand keeps recurring on this call, alongside a broader AI infrastructure thesis: global CSP capex up more than 60% in 2027, secured power capacity up over 100%, and an AI data centre infrastructure market compounding at roughly 40% annually to 2030. But look closely at the guidance, because the honest detail is buried there. Achievable demand growth for 2027 is put at mid-20% for DRAM and high-teens for NAND. Against a year in which revenue grew 328%, that is a dramatic deceleration in the rate of growth — and management is explicitly telling you it is a supply artefact, not a demand one. Customers' volume requirements are, in their words, more than double what the company can currently deliver. “Demand remains so strong that customers' volume requirement is more than double that of what we can supply today.” — Unknown Executive, Executive · 2026-10-27 Capital expenditure will rise significantly in 2027, with Yongin, M15X ramping to full capacity by year-end, and M17 in construction. But the sequencing is deliberately defensive: shells accelerate, cleanrooms open in phases. “More investment or more fabs does not necessarily mean more supply.” — Unknown Executive, Executive · 2026-10-27 It is a supply discipline argument dressed as a capex plan — a company pre-building optionality while refusing to commit bits until demand and pricing justify them.

The tape has already voted the other way

Here is the tension that makes this quarter worth reading rather than skimming. Every operating metric is accelerating, and the equity complex around it is being de-rated. Over the past 90 days, the market's AI data center theme has fallen 5.4%, with six constituents up and fifty down. High bandwidth memory exposure is down 7.2% across twelve tickers; data-centre capacity names are down 8%; even the hard disk drive complex has given back 6%. Semiconductor equipment, optics, and power infrastructure are all in the same drawdown bucket. That divergence explains two otherwise incongruous moments on the call. The CEO opens by thanking shareholders for their trust "amid the recent heightened market volatility," and the company confirms that Solidigm has abandoned its pre-IPO and IPO review — a financing decision reversed by market conditions rather than by the business. “After carefully considering recent financial market volatility and the correction in semiconductor stocks, Solidigm has concluded that an IPO would not fully deliver the anticipated benefits at this time.” — Noh-Jung Kwak, President and CEO · 2026-10-27 And then management does something public companies rarely do with this much directness: it tells you its own stock is mispriced, and puts KRW 20 trillion of this quarter's cash behind the claim.

We believe our share price today is undervalued, not adequately reflecting our earnings and cash flow. That is why we decided that share acquisition and cancellation would be an effective way of raising the per share value.
Unknown Executive, Executive · 2026-10-27

Two genuinely new threads — and one that quietly disappeared

The most interesting new demand vector is not HBM at all; it is NAND. As agentic AI and long-context workloads expand, the KV cache is outgrowing what HBM or DRAM can economically hold, which forces a tiered memory architecture. SK hynix is building SALT-KV, software that decides which cache lines live in HBM, DRAM or SSD, and CMM-hybrid, a CXL-based hybrid memory that parks the entire cache on SSD and pre-loads predictions into DRAM. “SSD is already being used in some AI systems to offload KV cache, and we believe its role will expand into the context memory tier.” — Unknown Executive, Executive · 2026-10-27 Longer term, HBF is aimed at the gap between HBM and SSD. This is the company riding a wave the market named two quarters ago — KV cache was already a top-ten global theme in the second quarter of 2026 — but with a proprietary implementation that turns NAND from a commodity into an AI systems component. The second thread is technological conservatism. 4F square DRAM appears in the top ten, but chiefly as a reminder that the company intends to sweat its 6F platform through EUV scaling before jumping, applying 4F or 3D DRAM only "when it is ready technologically and economically." An aggressive capex cycle run by a management team that refuses to rush node transitions is a different animal from the last two booms. What has quietly gone missing is the old bit-share-at-any-cost vocabulary. The emphasis now is high value mix, expanded HBM business, and the expectation that the profitability gap between HBM and conventional DRAM will narrow as HBM4 pricing is renegotiated for 2027. That is a subtle but important admission: the HBM premium is not going to widen forever, so the fix has to come from conventional DRAM pricing and cost — which is precisely what an 80% operating margin buys you room to do. The story, then, is not the record. It is the gap between a business that has pre-sold three years of output at 80% margins and a stock that management feels compelled to buy back in size. One of those two things is wrong. The risk to the bull case is not demand — it is the KRW 22.9 trillion reminder that a strategic equity stake can vaporise a third of a quarter's profit, and that a special dividend funded by investment gains is not a recurring line item.