SK hynix Sold Out to 2029: Allocation Buyer Guide

The short version

SK hynix's third quarter, reported October 28, was the most profitable in its history — and the most consequential for buyers who *don't* buy HBM. Revenue hit **104.62 trillion won** (roughly $75 billion at current exchange rates), the first quarter above 100 trillion, up 327.9% year on year and about 6.5% above the FnGuide consensus. Operating profit reached **83.33 trillion won**, up 632% year on year, at an **80% operating margin** (79.7% unrounded) — another record quarter. Then, on the earnings call, the company said the sentence that matters for everyone downstream: **its DRAM and HBM output is effectively sold out through 2029, counting signed long-term agreements (more than 50% of 2027-2028 capacity, with 14 customers) plus demand it is negotiating with customers.** Read that as a buyer, not an investor. One of the world's largest memory makers says its DRAM and HBM output is effectively sold out through 2029 — overwhelmingly to AI infrastructure customers on long-term agreements. Whatever conventional DRAM you plan to buy in 2027, 2028, or 2029 competes for the residual. Add two more signals from the same report: management expects the shortage to persist **through 2030**, and the company plans about **100 trillion won in 2026 shareholder returns** — dividends of at least 30 trillion won plus buybacks and cancellations of at least 60 trillion won, including the 40 trillion won buyback completed last week — while also planning a significant increase in capital spending next year as Yongin Phase 1 enters full operation. This article decodes the quarter for component buyers: what "sold out" actually covers, why it can coexist with forecasts of a 2028 price peak (#61), and the allocation playbook for the next three years.

The numbers, and what they say about pricing power

Start with the margin. An operating margin that climbed from ~72% in Q1 to 76.3% in Q2 to **79.7% in Q3** describes a seller with no need to negotiate. Memory has been cyclical for four decades; margins near 80% are what a market looks like when supply, not demand, sets every term. Revenue beat the FnGuide consensus by about 6.5% and operating profit by about 8.3% — in a quarter when analysts already expected records. Net profit was **39.78 trillion won**, up 216% year on year. Nothing in the operating business softened. The forward statements were stronger than the results. HBM4 — sixth-generation high-bandwidth memory — is expected to overtake HBM3E as the largest HBM line from Q4 (the company expects HBM4 sales to surpass HBM3E's), which means an even larger share of SK hynix's wafer starts will be committed to AI accelerators, where industry estimates put HBM at roughly three to four times the wafer capacity of conventional DRAM per bit (Micron: about 3x for HBM3E versus DDR5). And the shareholder-return plan — dividends of at least 30 trillion won plus buybacks and cancellations of at least 60 trillion won (including the 40 trillion won buyback completed last week), about 100 trillion won in total for 2026 — tells you how management ranks its options: record returns to owners, announced alongside a significant capex increase for next year rather than an all-out capacity race. None of this is criticism; it is simply the incentive structure your purchase orders now live inside.

"Sold out" vs. the 2028 glut forecast: both can be true

Attentive readers will spot the tension with #61, where analyst forecasts put a price peak around 2028 and a possible −23% DRAM correction as new capacity lands. If SK hynix is sold out through 2029, where would a glut come from? The resolution is that there are now **two memory markets**, and they are decoupling: - **The contract market** — HBM and LTA-committed conventional DRAM for hyperscalers and OEM giants — is what "sold out through 2029" describes. Volumes here are allocated years ahead, pricing is formula-bound, and availability for outsiders is extremely limited. - **The transactional market** — spot and short-cycle contract volume served by new conventional capacity (Chinese expansion, Korean and Micron fab ramps, the 1δ generation) — is where a 2028–29 correction would occur. Prices there can fall even while the contract market stays locked, because they clear different wafers for different buyers. The practical consequence: a future headline saying "DRAM prices fall 20%" may describe a market you cannot access at your volumes, while your allocated contract price barely moves — or the reverse, tightness persisting in your specific density long after generic benchmarks soften. Buyers who model one market will misread the next three years. The parts in our own catalog illustrate the split: a Samsung DDR4 like the K4A8G165WC-BCTD and a Micron DDR3L like the MT41K128M16JT-125 live entirely in the transactional world — their pricing will follow capacity and channel stocks, not SK hynix's allocation book. Know which market each line of your BOM actually buys from.

What this means for component buyers

**1. Treat allocation as the default, not the exception, through 2029.** When one of the world's largest memory makers is effectively sold out through 2029, Samsung and Micron face the same demand queue — expect parallel sold-out language and tightening LTA terms across all three. If your 2027–29 DRAM volumes matter, the time to secure framework allocation was before this earnings call; the second-best time is now, while suppliers are still signing. **2. Stop optimizing spot price; start optimizing certainty.** At 79.7% supplier margins, spot-hunting saves basis points and risks line-down quarters. Re-rank your KPIs: coverage months, allocation confirmation, and second-source qualification status are worth more than the last 3% on a spot buy in a sold-out market. **3. Read capital returns as a supply forecast.** 100 trillion won returned to shareholders is capital not building fabs — though SK hynix also plans a significant capex increase next year. Across the industry, buyback-and-dividend discipline at peak margins means, in our view, the capacity response to this shortage is likely to be slower than in past cycles — one reason management can credibly guide shortage "through 2030." Build that slower response into your own supply assumptions. **4. Watch the HBM wafer tax on conventional supply.** HBM4 overtaking HBM3E as the largest HBM line from Q4 tightens conventional DRAM mechanically: every wafer diverted to HBM (at roughly 3-4x wafer capacity per bit) subtracts conventional bits. Even flat total wafer capacity means shrinking conventional output share. Densities and product lines that compete most directly with HBM for wafer starts face the tightest residual supply. **5. Use the two-market split in negotiations and design.** For transactional-market parts, keep flexibility: multi-source footprints, density agility, and channel partners with real allocation relationships. For anything drifting toward the contract market — server RDIMMs, high-density modules — assume LTA dynamics apply to you earlier than you expect, and negotiate escalators and floors with the 2028 scenarios from #61 on the table, not as surprises. The DRAM lines we carry — Samsung's K4A8G165WC-BCTD DDR4, Micron's MT41K128M16JT-125 DDR3L, and the module ecosystem around them — are transactional-market parts, and we track their channel availability weekly. Send us your 2027–29 memory requirements and we'll map them against the allocation reality: what can still be covered transactionally, what needs framework cover, and where a design change buys you out of the tightest queues.

Frequently asked questions

**What exactly did SK hynix say was sold out?** On its October 28 earnings call, the company said its DRAM and HBM output is effectively sold out through 2029, counting signed long-term agreements (more than 50% of 2027-2028 capacity, with 14 customers) plus demand it is negotiating with customers — predominantly long-term deals with AI infrastructure buyers. It also guided that the memory shortage should persist through 2030. **Does "sold out" mean I can't buy SK hynix DRAM until 2030?** No. It means the factory's output is committed or under negotiation — distributors and OEMs holding those commitments still sell parts, and transactional volume exists. What changes is leverage and price: you buy from the allocation holders, at scarcity prices, with limited volume flexibility. "Sold out" describes who holds the power, not a literal empty shelf. **How can there be a 2028 glut forecast (#61) if capacity is sold out to 2029?** Because the two describe different pools. The sold-out pool is contract/HBM capacity. The forecast correction concerns transactional conventional DRAM, where new Chinese capacity, fab ramps, and Micron's 1δ generation add bits. Contract and spot prices can move in opposite directions; plan each BOM line against the market it actually buys from. **Why do shareholder returns matter to chip buyers?** Capital allocation is a supply signal. Returning ~100 trillion won in 2026 while guiding shortage through 2030 says management is harvesting scarcity even as it plans a significant capex increase next year. Multiply that discipline across vendors and the industry's supply response slows — extending the allocation era for buyers. **Samsung reports tomorrow — will it say the same?** Samsung pre-announced record preliminary results on October 8, with third-quarter operating profit of about 107.4 trillion won, up roughly 783% (nearly nine times) from a year earlier. Watch its call for the same three tells: LTA/sold-out language, HBM4 mix, and capex-versus-returns discipline. If Samsung echoes "sold out," the allocation era is industry-wide fact, not one vendor's posture. We'll cover it either way. **What should a mid-sized buyer do this quarter?** Three moves: (1) quantify your 2027–29 DRAM exposure by line and market (contract-exposed vs transactional); (2) open framework/allocation discussions with suppliers and channel partners now, accepting price formulas in exchange for volume certainty; (3) qualify second sources and flexible footprints for your highest-volume densities while qualification is still a choice, not an emergency. ---

Sources

Use the latest guidance for the exact part and manufacturer when deciding acceptance criteria.