SK hynix Q3 2026 Earnings Explained: Record ₩104.6T Revenue, HBM4 Ramp and Long-Term Supply Deals
SK hynix reported Q3 2026 (July–September) consolidated revenue of KRW 104.6155 trillion, operating profit of KRW 83.3338 trillion (80% operating margin) and net profit of KRW 39.7756 trillion on October 28, 2026. According to the company, this is the first quarter in its history with revenue above KRW 100 trillion. The drivers it named were HBM4 volume shipments, server DDR5 and 321-layer TLC enterprise SSDs, and it expects HBM4 to overtake HBM3E in both shipments and revenue in Q4. Below: the numbers, why net profit fell while operating profit rose, what the HBM4 ramp means technically, and how long-term supply agreements change the way memory is sold.
Not investment advice; no share price or target price discussion. All figures are as reported by the company (K-IFRS, consolidated, preliminary and unaudited), taken from the SK hynix newsroom release and the DART preliminary earnings filing. Sources checked on October 28, 2026.
SK hynix Q3 2026 earnings at a glance
One-line answer: Revenue rose 32% quarter over quarter and 328% year over year; operating profit rose 38% QoQ and 632% YoY. Both are company records.
| KRW trillion | Q3 2026 | Q2 2026 | QoQ | Q3 2025 | YoY |
|---|---|---|---|---|---|
| Revenue | 104.62 | 79.32 | +32% | 24.45 | +328% |
| Operating profit | 83.33 | 60.54 | +38% | 11.38 | +632% |
| Operating margin | 80% | 76% | +3%p | 47% | +33%p |
| Net profit | 39.78 | 93.92 | -58% | 12.60 | +216% |
Figures are rounded from the company’s comparison table. Per the DART filing, nine-month revenue reached KRW 236.51 trillion and nine-month operating profit KRW 181.49 trillion. At quarter end, cash and cash equivalents were KRW 114.5 trillion (up 26.5 trillion QoQ), borrowings KRW 15.8 trillion (down 2.8 trillion), and net cash KRW 98.7 trillion.
Why did SK hynix’s net profit drop in Q3 2026?
The gap is outside operations. In the DART filing, Q3 pre-tax income from continuing operations was about KRW 50.71 trillion, well below operating profit of KRW 83.33 trillion. In Q2 the opposite happened: pre-tax income (about KRW 122.71 trillion) and net profit (KRW 93.92 trillion) were both larger than operating profit (KRW 60.54 trillion). So the 58% QoQ fall in net profit comes from non-operating items, not from the core memory business. The press release does not break down what those items were, so this post does not guess. For the underlying business trend, revenue and operating profit are the cleaner signals.
What drove the quarter?
One-line answer: Higher memory prices from AI demand, plus a product mix shifting toward data-center parts: HBM4, server DDR5 and eSSD.
- Pricing: the company says prices of key products rose as AI memory demand grew.
- DRAM: HBM4 shipments began in earnest and server DDR5 sales rose significantly.
- NAND: sales of 321-layer TLC enterprise SSDs expanded, along with other data-center products.
The release does not disclose DRAM vs. NAND revenue or the HBM share of revenue, so no segment split is estimated here.
What does “HBM4 becomes the mainstream product in Q4” mean?
One-line answer: The main HBM generation shipping next to AI accelerators moves from HBM3E to HBM4, and SK hynix is pushing its M15X fab to full capacity by year end to make that happen.
HBM stacks several DRAM dies, links them with through-silicon vias (TSVs), and sits them on a base (logic) die. The stack is packaged right next to a GPU or ASIC and talks to it over a very wide bus. The JEDEC HBM4 standard (JESD270-4, announced April 16, 2025) doubles the interface to 2048 bits (HBM3E uses 1024), doubles independent channels to 32, targets up to 8 Gb/s per pin and up to 2 TB/s per stack, and supports 4-, 8-, 12- and 16-high stacks with 24 Gb or 32 Gb dies.
What SK hynix said about HBM this quarter:
- HBM4 shipments began in earnest in Q3.
- M15X will be ramped to maximum capacity by the end of the year to increase HBM4 output.
- In Q4, HBM4 is expected to surpass HBM3E in both shipment volume and revenue.
From an engineering view, HBM supply is hard because every stack starts as a DRAM wafer. The dies have to be made, stacked, tested with the base die and then co-packaged with the accelerator (2.5D packaging such as TSMC CoWoS) before a server ever sees them. Each step adds yield loss and lead time, and every wafer used for HBM is a wafer not sold as standard DRAM. Packaging is covered separately in the SK hynix HBM5 and TSMC CoWoS post, and a competitor’s custom HBM4 approach in the Micron HBM4 NVHBM post.
Server DDR5, SOCAMM and the 1c node
One-line answer: AI servers need more than HBM. The CPU side needs large DDR5 and LPDDR pools too, and the 1c (sixth-generation 10nm-class) node is how SK hynix adds bits without waiting for new fabs.
- 1c node: the company plans to raise the 1c share quickly. A denser node yields more bits per wafer from the same cleanroom.
- LPDDR5 SOCAMM: a detachable server module built from low-power LPDDR5 chips; the company plans to grow sales for AI servers.
- Server DDR5: sales rose significantly in Q3 and remain a growth target.
Where do NAND and Solidigm fit?
One-line answer: NAND growth is led by 321-layer TLC eSSDs for data centers, and the planned Solidigm IPO review has been halted.
SK hynix’s NAND business combines its Korean NAND fabs with Solidigm, its US enterprise-SSD subsidiary. In Q4 the company will raise the 321-layer node to more than half of its domestic NAND capacity to supply more high-performance, high-capacity eSSDs. Longer term it wants to use HBF (high bandwidth flash) and system-level optimization to make NAND a lever for AI service throughput and cost.
On Solidigm, CEO Kwak Noh-jung said the IPO was meant to fund Solidigm’s capacity with outside capital and ease SK hynix’s own investment burden, but that given recent financial market conditions the expected benefits could not be fully achieved now, so the review was stopped.
How do SK hynix’s long-term agreements (LTAs) change memory supply?
One-line answer: More than half of 2027 and 2028 capacity is already committed to 14 customers, with prices that move within pre-agreed ranges.
| Item | What the company disclosed |
|---|---|
| Customers | 14 major customers, including global hyperscale cloud providers |
| Volume | More than 50% of total 2027 and 2028 production capacity |
| Term | Typically five years; extensions under discussion with some customers |
| Pricing | Reflects market conditions within a price range agreed with each customer |
Memory has historically been a boom-and-bust business. Locking in a majority of future output gives SK hynix demand visibility, which matters because fabs and tools take years from decision to output. For buyers, it means memory is increasingly allocated through multi-year contracts rather than bought on short notice.
Outlook and capex: what SK hynix guided
One-line answer: The company expects demand to exceed supply through 2030 and plans significantly higher investment next year, led by the first Yongin fab; it gave no capex figure in the release.
- Demand: enterprise and personal AI in the near term, physical AI (autonomous driving, robotics) longer term, with AI spending broadening from a few companies to enterprises and governments.
- Supply: leading-edge process difficulty and long fab lead times keep supply behind demand through 2030, per the company.
- Capex: the first Yongin fab begins full operation next year and investment will rise significantly versus this year, with size and timing adjusted to demand, competitiveness, profitability and efficiency.
- Direction: grow revenue and profit another step next year.
Shareholder returns were also announced: roughly KRW 100 trillion for fiscal 2026 (cash dividends of KRW 30 trillion or more, buybacks and cancellations of KRW 60 trillion or more), and a Q3 dividend of KRW 20,000 per share, half of it a special dividend from realized Kioxia investment gains.
What engineers should take from this report
- HBM generation timing: if HBM4 overtakes HBM3E in Q4 as guided, accelerator designs shipping from 2027 will assume the 2048-bit HBM4 interface.
- Host memory matters: the server DDR5 and SOCAMM focus shows CPU-side memory is becoming an AI infrastructure constraint too.
- Storage tier: 321-layer eSSDs and HBF point to NAND being treated as part of AI serving throughput, not just capacity.
- Allocation: with over half of 2027–2028 capacity under LTAs, memory procurement is moving toward long-term contracts.
FAQ
What was SK hynix’s Q3 2026 revenue and operating profit?
KRW 104.6155 trillion in revenue and KRW 83.3338 trillion in operating profit (80% margin), per the company’s preliminary K-IFRS consolidated results.
Did SK hynix disclose its HBM revenue share for Q3 2026?
No. The press release gives no HBM share and no DRAM/NAND revenue split. It only says HBM4 should exceed HBM3E in shipments and revenue in Q4.
How much will SK hynix spend on capex in 2027?
No figure was given in the release. The company said investment will rise significantly from this year, with scale and timing kept flexible.
Bottom line
SK hynix’s Q3 2026 was a record quarter by its own numbers: KRW 104.6 trillion in revenue and KRW 83.3 trillion in operating profit. The technical story is the HBM3E-to-HBM4 handover in Q4, the rise of server DDR5, SOCAMM and 321-layer eSSDs as high-value products, and LTAs that commit over half of 2027–2028 capacity to 14 customers. The drop in net profit came from non-operating items that the release does not explain.
Sources
- SK hynix Newsroom — SK hynix Announces 3Q26 Financial Results (October 28, 2026)
- SK hynix Newsroom (Korean) — 2026년 3분기 경영실적 발표
- DART — SK hynix preliminary consolidated earnings filing (October 28, 2026, 08:28 KST)
- SK hynix IR — Earnings Release (FY2026 Q3)
- JEDEC — JESD270-4 HBM4 Standard (April 16, 2025)