What changed #
On 29 July 2026 SK hynix reported its Q2 2026 results (quarter ended 30 June 2026), and they set new company records on every headline measure:
- Revenue: 79.32 trillion won (about $58 billion), up 257% year over year and above the prior quarter's 52.58 trillion won.
- Operating profit: 60.54 trillion won, a 76% operating margin and a 557% increase versus Q2 2025.
- Net profit: 93.92 trillion won, a 118% net margin.
- First-half revenue crossed 100 trillion won for the first time in the company's history.
The driver was not a single product but a stack of high-value lines: HBM, AI-server DRAM and enterprise SSDs (eSSD). SK hynix said both DRAM and NAND contract prices rose "significantly" quarter over quarter, and that HBM4 — its next-generation stack — began mass shipments during the quarter, with HBM4E samples already shipped in the first half.
A brokerage review of the results (First Shanghai, 4 Aug 2026) breaks the quarter down further: DRAM revenue about 58.4 trillion won (+241% YoY), NAND about 20.9 trillion won (+342% YoY), and HBM revenue about 9.4 trillion won (+31% YoY). On pricing, the firm cites DRAM ASP up roughly 30% QoQ and NAND ASP up more than 50% QoQ, based on the earnings call.
Why buyers should care #
Three things in this report matter directly to component procurement, beyond the headline profit numbers:
1. Supplier pricing power is structural, not temporary. A 76% operating margin in memory is exceptional. SK hynix attributed it to constrained supply of high-value parts and said cash reached 88 trillion won with net cash of 69.4 trillion won. Suppliers in this position do not need to discount to fill capacity; they allocate it. For buyers, that means contract pricing is more likely to rise than to soften through 2027.
2. HBM4 is now a shipping product, but volume is committed. Mass shipments of HBM4 starting in Q2 is a real milestone, yet SK hynix also disclosed it has finalized long-term agreements (LTAs) with around 10 customers. When the newest, highest-margin stack is already tied to multi-year contracts, the free spot market for HBM4 and even HBM3E stays thin. Buyers not inside an LTA should expect tight allocation rather than open availability.
3. Conventional DRAM and NAND are being squeezed by HBM. HBM consumes far more wafer area per bit than standard DRAM. As SK hynix and peers shift capacity toward HBM and AI-server DRAM, the available output for commodity DDR4/DDR5 and NAND tightens. That is the mechanism behind the 30%+ / 50%+ QoQ price moves the earnings call described — and why those moves are likely to persist.
Affected component areas #
- HBM — HBM4 in mass production, HBM4E sampling; allocation already committed to lead customers.
- DRAM (DDR4/DDR5, LPDDR, server modules) — ASP up ~30% QoQ; capacity redirected toward AI-server and HBM-adjacent output.
- NAND flash / eSSD — ASP up 50%+ QoQ; enterprise SSD demand roughly doubling per the earnings commentary.
- AI-server platforms — the primary end demand pulling every memory tier.
- Automotive and industrial memory — less directly inflated, but exposed to the same DRAM/NAND capacity competition.
HBM4 and the allocation shift #
The most consequential combination in the release for buyers is "HBM4 mass shipments began" together with "around 10 LTAs finalized." It tells us two things: the leading-edge stack is now real and shipping, and its output for the next several quarters is largely pre-sold.
This is consistent with SK hynix's separate August 2026 fab commitment — 54 trillion won for Yongin Y2 (DRAM/HBM) and Cheongju M17 (NAND), ramping through 2031. Capacity is being added, but on a multi-year horizon, not in time to relieve 2026–2027 tightness. Our earlier note on reported no-cap LTA pricing covers what those agreements may mean for contract structure.
RFQ and sourcing checks #
| Risk signal | Buyer action |
|---|---|
| DRAM/NAND prices up 30–50%+ QoQ | Request a dated quotation for the exact orderable code before releasing POs |
| HBM4/HBM3E allocation committed to LTAs | Confirm whether your volume is inside an approved agreement or on allocation |
| Capacity shifting to HBM | Lock multi-quarter coverage for DDR4/DDR5 and eSSD where the BOM allows |
| Tight spot stock | Verify package, date code and lot traceability on every delivery |
| New-node DRAM (1c) ramping | Qualify second-source equivalents early; do not assume drop-in substitution |
Practical takeaway #
SK hynix's Q2 2026 results confirm that the memory up-cycle is being driven by AI infrastructure, not by a cyclical snap-back. Record margins, HBM4 mass shipments and ~10 signed LTAs point to continued pricing power and tight allocation into 2027. Buyers should treat 2026–2027 memory as allocation-constrained: confirm current pricing and date codes on every order, secure multi-quarter coverage for critical DDR4/DDR5 and NAND lines, and verify whether HBM requirements are inside or outside an existing supply agreement before committing design or production plans.
Sources #
- SK hynix, "SK hynix Announces 2Q26 Financial Results," news.skhynix.com, 29 July 2026.
- First Shanghai Securities, SK hynix Q2 2026 earnings review, 4 August 2026 (segment revenue and ASP movements cited from the earnings call).
- LimChip prior coverage: SK hynix LTA pricing (2026-07-17), Yongin/Cheongju fab capex (2026-08-09).
Need stock, date-code or package confirmation?
Send the part number, quantity, target date code and packaging requirements. LimChip will check available lots and RFQ details before you place the order.
Send RFQ