PER, EPS, PBR and BPS are calculated in-house from the last 4 quarters (2025Q3–2026Q2) · Dividend yield is based on ₩3,000 per share · Prices as of the 2026-10-02 close
Quarterly results keep setting record highs, yet market attention has already shifted to when memory prices peak and how fast rivals ramp HBM4.
Second-quarter 2026 revenue of KRW 79.32tn and operating profit of KRW 60.54tn set fresh quarterly records, and first-half revenue topped KRW 100tn for the first time.
Second-quarter net profit attributable to owners of KRW 93.82tn included large gains on investment asset disposals and revaluation, so it should be read separately from operating performance.
HBM4 began mass-production shipments in the second quarter of 2026 with volume expansion slated for the second half, while HBM4E has completed sampling and targets mass production in 2027.
On August 19 the board approved about KRW 40tn of share buybacks for cancellation and changed the payout standard to 'at least 50%' of cumulative free cash flow, with further measures to be detailed at the third-quarter results release.
Despite the records, second-quarter operating profit came in below consensus, and some brokers cut target prices citing falling NAND prices and rivals' HBM4 expansion, leaving bullish and bearish views coexisting.
SK hynix is a pure-play memory maker centered on DRAM and NAND flash, with Icheon and Cheongju as core domestic production hubs and advanced packaging infrastructure being expanded alongside.
Its portfolio spans HBM for AI accelerators, high-capacity server DRAM modules, enterprise SSDs and mobile NAND, plus image sensors; in its second-quarter commentary the company said it generated its best-ever profitability by expanding sales of high-value products such as HBM, AI server DRAM and eSSD.
In HBM, Counterpoint Research data show a 62% shipment share in the second quarter of 2025 for the top position, and 57% on a third-quarter revenue basis, indicating an oligopolistic standing.
Its customer base is broadening from AI accelerator vendors including Nvidia to large cloud providers designing their own chips, and UBS analyzed that SK hynix would be the first HBM3E supplier for Google's latest TPU v7p and v7e.
On demand visibility, the company said it wrapped up long-term agreement negotiations with about ten customers including key accounts and continues additional discussions.
In commodity DRAM and NAND, SOCAMM2 sales rose sharply in the second quarter and supply of products on the sixth-generation 10nm-class (1c) node began in earnest, while 321-layer NAND already accounts for the largest share of output and is planned to reach about 50% of domestic capacity by year-end.
The competitive structure remains a three-way contest with Samsung Electronics and Micron, and Samsung declared the industry's first HBM4 mass-production shipment in February 2026, reshaping the share battle through the generational transition.
Management framed its direction around the view that memory competitiveness now extends into system architecture and packaging, so it intends to lead system-level memory innovation on the back of a broad portfolio and co-development capability with customers.
| Quarter | Revenue | Operating profit | Op. margin |
|---|---|---|---|
| 2025Q2 | ₩22.2T | ₩9.2T | 41.4% |
| 2025Q3 | ₩24.4T | ₩11.4T | 46.6% |
| 2025Q4 | ₩32.8T | ₩19.2T | 58.4% |
| 2026Q1 | ₩52.6T | ₩37.6T | 71.5% |
| 2026Q2 | ₩79.3T | ₩60.5T | 76.3% |
| Year | Revenue | Operating profit | Net income (parent) | Op. margin | ROE | Debt-to-equity |
|---|---|---|---|---|---|---|
| 2022 | ₩44.6T | ₩6.8T | ₩2.2T | 15.3% | 3.5% | 64.1% |
| 2023 | ₩32.8T | -₩7.7T | -₩9.1T | −23.6% | −17.0% | 87.5% |
| 2024 | ₩66.2T | ₩23.5T | ₩19.8T | 35.5% | 26.8% | 62.2% |
| 2025 | ₩97.1T | ₩47.2T | ₩42.9T | 48.6% | 35.6% | 45.9% |
Consolidated basis (financial statements including subsidiaries) · Final figures from DART filings · Net income attributable to owners of the parent · Data as of 2026-09-04
The confirmed numbers show the full amplitude of the memory cycle.
From 2022 revenue of KRW 44.62tn and operating profit of KRW 6.81tn (15.3% margin), the company swung to a 2023 operating loss of KRW 7.73tn on revenue of KRW 32.77tn, then returned to profit in 2024 with revenue of KRW 66.19tn and operating profit of KRW 23.47tn (35.5%), before expanding to KRW 97.15tn in revenue and KRW 47.21tn in operating profit (48.6%) in 2025.
The balance sheet improved alongside: the debt-to-equity ratio fell from 87.5% in 2023 to 62.2% in 2024 and 45.9% in 2025, while operating cash flow rose from KRW 4.28tn in 2023 to KRW 53.37tn in 2025.
Quarterly, results expanded for five straight periods: KRW 22.33tn revenue and KRW 9.21tn operating profit in the second quarter of 2025, then KRW 24.45tn/KRW 11.38tn, KRW 32.83tn/KRW 19.17tn, KRW 52.58tn/KRW 37.61tn, and KRW 79.32tn/KRW 60.54tn in the second quarter of 2026.
First-half 2026 revenue alone exceeded full-year 2025 revenue, and cumulative first-half revenue passed the KRW 100tn mark for the first time ever. On margins, the company said the second-quarter operating margin improved 5 percentage points quarter on quarter to 76%, a record for both operating profit and margin.
Second-quarter net profit attributable to owners of KRW 93.82tn far exceeded operating profit because of non-operating items: management explained that KRW 1.1tn of foreign exchange gains and KRW 63.3tn of gains on investment asset disposals and revaluation drove net non-operating income of KRW 62.2tn.
In other words, a large share of second-quarter net profit is low-recurrence in nature, a point to keep in mind when comparing with first-quarter 2026 net profit of KRW 40.33tn.
Even so, despite the records, both revenue and operating profit came in below the FnGuide consensus of KRW 83.9tn in revenue and KRW 64tn in operating profit, and the company said the shift of some high-value shipments into the second half and portfolio composition appeared to affect blended average selling prices.
On liquidity, cash and cash equivalents stood at KRW 88tn at end-June with borrowings of KRW 18.6tn, strengthening its net cash position.
This cycle is defined by a supplier-favored phase created by AI infrastructure investment. The World Semiconductor Trade Statistics organization projected the 2026 global semiconductor market to grow more than 25% year on year to about USD 975bn, with memory rising in the 30% range, above the overall growth rate.
Prices did surge: TrendForce data show global DRAM industry revenue of USD 97bn in the first quarter of 2026, up 81% quarter on quarter, with commodity DRAM contract prices up 93-98% sequentially. The same firm expected commodity DRAM contract prices to climb a further 58-63% quarter on quarter in the second quarter.
On supply, HBM output continues to absorb wafer capacity and push commodity products down the priority list, with assessments that SK hynix's HBM, commodity DRAM and NAND capacity is effectively sold out through 2026.
A peak debate runs in parallel, however: Raymond James analyst Karl Ackerman raised the possibility that DRAM and NAND average selling prices peak around mid-2026, while also noting that long-term supply agreements with AI customers could cushion downward price pressure.
Competitively, the generational shift is the swing factor. Goldman Sachs analyzed that SK hynix would keep a dominant position in HBM3 and HBM3E and hold above 50% of the total HBM market at least through 2026, and UBS expected roughly a 70% share in the HBM4 market for Nvidia's next-generation Rubin platform.
Among latecomers, the Korea-China technology gap is estimated at two to three generations while CXMT plans to channel IPO proceeds into expanding HBM lines, leaving a medium-term variable in place.
Management's stated second-half direction is simultaneous improvement in volume and mix.
It said HBM4 volumes would expand in earnest and 1c-node commodity DRAM shipments would increase, so second-half bit growth should exceed the first half, and that wider HBM4 sales and a higher share of high-value products should contribute positively to blended average selling prices.
On the product calendar, HBM4 mass-production shipments began in the second quarter with output scaling up in the second half, while HBM4E has completed customer sampling and is being developed for full mass production in 2027.
On capacity, the company said it is pulling forward the M15X mass-production schedule and investing so that capacity can be expanded quickly after the Yongin phase-one fab cleanroom opens in early 2027, with 2026 investment expected in the high KRW 40tn range.
It also plans the P&T7 advanced packaging plant, the new M17 NAND site, and an additional domestic cluster beyond Yongin, while maintaining the principle that fab construction, equipment installation and capacity additions will proceed in stages based on customer demand visibility and investment efficiency.
On capital allocation, the board on August 19 approved treasury share purchases totaling KRW 40.0043tn, to be bought on-market for about three months from August 20 and cancelled in full upon completion.
It further stated that the payout standard was changed from 'within 50%' to 'at least 50%' of cumulative free cash flow, with the specific size and method of additional returns to be announced at the third-quarter results release after board approval.
On demand durability, the company argued that infrastructure investment aimed at not falling behind in the AI race is likely to remain solid.
After a sharp jump in profits over the last four reported quarters, the share price multiple measured against earnings sits toward the lower part of the historical trading band typically seen during memory upcycles.
That earnings base, however, contains sizeable low-recurrence items such as the KRW 63.3tn of gains on investment asset disposals and revaluation booked in the second quarter, so an earnings multiple alone is a weak gauge of level.
Against net assets the stock trades at a premium, and the multiple differs widely depending on whether book value per share is taken from in-house calculations or from the exchange's disclosed figure, which itself shapes interpretation.
On dividend yield alone the level runs below the market average, but shareholder returns are weighted more toward share-count reduction, as with the KRW 40tn buyback to be cancelled in full, and the size and method of additional returns are due to be announced at the third-quarter results release.
Because cyclical earnings make multiples look compressed near profit peaks and stretched at troughs, the durability of earnings rather than the absolute multiple remains the key variable.
PER, EPS, PBR and BPS are all calculated in-house (the same method as Naver and Toss) · Dividend yield = cash dividend per share from DART filings (supplemented by KRX) ÷ current price · As of 2026-09-04
Counterpoint Research data confirmed a 62% HBM shipment share and the top position in the second quarter of 2025, and Goldman Sachs projected continued HBM3/HBM3E dominance and an above-50% share at least through 2026.
On the generational shift, the company said HBM4 yield and quality have risen close to prior-generation levels and are supporting stable supply. UBS also expected roughly a 70% share in the HBM4 market for the Rubin platform.
The company said it concluded long-term supply agreement talks with about ten customers including key accounts and continues further discussions. On its first-quarter call it noted that customer-requested demand over the next three years already far exceeds its supply capacity.
For any price-decline phase, there is a market view that long-term contracts could cushion downward pressure. Whether that contract structure dampens earnings volatility is the point to watch.
On confirmed figures, operating cash flow expanded from KRW 4.28tn in 2023 to KRW 53.37tn in 2025, while the debt-to-equity ratio fell from 87.5% to 45.9% over the same span. The company said net cash stood at about KRW 69tn at end-June, reflecting sharply improved cash generation.
On that basis, a KRW 40tn treasury share purchase with full cancellation afterward was approved, and the payout standard was raised to at least 50% of cumulative free cash flow.
Despite the records, second-quarter revenue and operating profit both came in below the FnGuide consensus of KRW 83.9tn and KRW 64tn.
On the price cycle, there is a view that DRAM and NAND average selling prices may peak around mid-2026, and in the memory industry a price peak is usually read as a signal of rising supply and slowing conditions. How the market responds when the earnings slope flattens is the item to monitor.
Competition in the generational transition intensified after Samsung Electronics declared the industry's first HBM4 mass-production shipment in February 2026.
LS Securities on August 31, 2026 lowered its SK hynix target price from KRW 3.3mn to KRW 2.4mn, saying it reflected the possibility that the premium arising from supplier concentration would ease as rivals expand HBM4 supply.
The same report explained that this was not a call for weaker HBM demand or the end of the cycle, but a downgrade of expected excess profitability. Retaining share within key customers on next-generation products remains the central issue.
Lee Min-hee of BNK Investment & Securities assessed that NAND spot prices have trended down since late April amid weaker consumer demand, that doubts over hyperscalers' AI capex funding are growing, and that concerns about a swing to oversupply are widening as manufacturers announce large expansions.
Target price revisions followed: Mirae Asset Securities cut its SK hynix target from KRW 4.2mn to KRW 2.8mn on July 29, 2026, and BNK Investment & Securities lowered its target from KRW 1.85mn to KRW 1.48mn the same day.
Memory earnings are highly price-elastic. The confirmed record already shows a full reversal from a KRW 7.73tn operating loss in 2023 to KRW 47.21tn of operating profit in 2025.
Current results reflect the benefit of a phase in which first-quarter commodity DRAM contract prices rose 93-98% quarter on quarter, so a slowdown or reversal in contract pricing would hit margins quickly. With a growing depreciation base from heavy capex, a price downturn could amplify the downside.
In the HBM4 generation, yield, performance and supply reliability are contested simultaneously.
The company argued that HBM competitiveness is complete only when it spans meeting customer performance requirements and mass-supplying at stable yield and quality, but share can swing quarter to quarter depending on customer qualification timing and ramp speed.
From China, an estimated two-to-three-generation gap and CXMT's plan to expand HBM lines with IPO proceeds remain medium-term variables. Any delay in the generational shift or early yield trouble would directly disrupt the high-value mix improvement plan.
There are notes that the US Trade Representative opened Section 301 investigations into 16 countries including Korea in March 2026, raising uncertainty around export conditions to the United States.
On capital allocation, the plan for 2026 investment in the high KRW 40tn range is running alongside a KRW 40tn buyback-and-cancellation program, so a downturn could force a re-prioritization between investment and returns.
The company itself said fab construction, equipment installation and capacity expansion will proceed in stages considering customer demand visibility and investment efficiency. In this phase, any change in execution pace versus plan can itself be read as an industry signal.
Alongside third-quarter revenue and operating profit, the specific size and method of additional shareholder returns are due to be announced after board approval. The key is whether the guided second-half bit growth above first-half levels shows up in the actual numbers.
Since the on-market purchase runs about three months from August 20 with full cancellation to follow, the actual quantity and amount acquired and the cancellation filing need checking. A lower share count feeds directly into per-share metrics.
Time to check whether wider HBM4 shipments actually lift blended average selling prices, and whether the plan to raise 321-layer NAND to about 50% of domestic capacity by year-end is executed.
Whether the Yongin phase-one fab cleanroom opening and the accelerated M15X mass-production schedule proceed as planned is the inflection point for supply capability. It is also worth checking how execution pace is reconciled with the company's stated capex discipline.
Whether the HBM4E target of full mass production in 2027 holds, and how rivals' HBM4 ramp shifts customer-level share, will determine the outcome of the generational contest.
On confirmed figures, SK hynix moved from a 2023 operating loss through a sharp profit recovery in 2024-2025 into a phase where first-half 2026 revenue alone exceeded full-year 2025 revenue.
In the second quarter of 2026 it again set quarterly records with KRW 79.32tn in revenue and KRW 60.54tn in operating profit, and the company said the 76% operating margin was also an all-time high.
Net profit, however, included KRW 63.3tn of gains on investment asset disposals and revaluation, which are low-recurrence, and operating profit fell short of consensus.
The bullish case rests on the oligopolistic HBM position, long-term supply agreements with about ten customers, and capital allocation running from roughly KRW 69tn of net cash at end-June into a KRW 40tn buyback-and-cancellation.
The bearish case centers on the possibility of a mid-2026 price peak, falling NAND prices, and the chance that the supplier premium eases as rivals expand HBM4 supply.
The axes to watch are therefore the second-half HBM4 ramp and blended average selling prices, the size and method of additional returns to be disclosed at the third-quarter results release, and actual execution of expansion schedules such as Yongin and M15X.
This report is for information purposes only and contains no buy or sell opinion or target price for any security.
Report written 2026-09-05 · Data as of 2026-09-04
This content is AI analysis of market data and web search results, provided for information only. It is not a solicitation or recommendation to invest. Investment decisions and their consequences are the investor's own responsibility. Data may be delayed or contain errors.