Fundamental research · SK hynix · KRX: 000660 · Ordinary shares
Analysis and valuation date: October 15, 2026. Information cutoff: 20:25 KST. Reference: Korean regular-session close of ₩1,262,000, down 11.06%. Financial amounts are KRW trillion unless indicated; scenario values are won per Korean ordinary share. The financial baseline is the company’s July Q2 presentation, explicitly marked preliminary. Q3 results were still ahead. [13]
Our view: value depends on the earnings that survive the cycle
SK hynix sits at an important point in AI infrastructure: processors need fast local memory, servers need system memory, and data centers need persistent storage. Its opportunity spans all three. The valuation question is how much of today’s exceptional pricing and profitability can persist as customers optimize spending and competing capacity becomes available.
Our initial 12-month Base value is ₩1,857,000, with Bear ₩713,000 and Bull ₩3,018,000. The horizon is approximately October 2027 using FY2027 earnings. These are conditional market-valuation scenarios without assigned probabilities. The Base implies 47.1% appreciation from the reference close; dividends are excluded. The Bear allows substantial losses even after the recent selloff.
| Scenario | 12-month value | Vs. reference close |
|---|---|---|
| Bear | ₩713,000 | -43.5% |
| Base | ₩1,857,000 | +47.1% |
| Bull | ₩3,018,000 | +139.1% |
Our Base is 40.9% below the accessed 39-analyst target mean. That difference triggers an assumption review: our earnings forecast is materially lower than the displayed consensus. We explain both the gap and the model’s limitations below. A large forecast spread is not proof that either side is correct.
Three businesses share the AI opportunity, with different economics
| Business | Economic role | What investors should test |
|---|---|---|
| HBM | High-bandwidth memory close to an accelerator | Customer qualification, package yield, supply allocation and price per usable stack |
| Server and other DRAM | Working memory across computing systems | Bit shipments, contract prices, product mix and manufacturing cost |
| NAND and enterprise SSDs | Persistent storage, including Solidigm products | Enterprise mix, controller value, cost per bit and storage pricing |
HBM can create differentiation through engineering and reliable delivery. A demanding product that is qualified inside an accelerator can be harder to substitute than a standardized memory component. The supplier must still earn a return after manufacturing losses, packaging work and continuing development. Product relevance alone does not establish the margin on every generation.
The conventional memory cycle remains important. Rising prices can increase revenue even without an equivalent increase in physical output. When supply tightness eases, a company can ship more bits while earning less on each bit. For this reason, we track volume and pricing separately and avoid treating revenue growth as a clean measure of AI adoption.
NAND adds another route to AI spending through enterprise storage. Its pricing cycle can differ from HBM, and its capital demands compete with DRAM investment. A stronger SSD portfolio may improve mix, but does not make flash memory immune to excess supply. The consolidated investment case needs all of these businesses to earn their cost of capital.
The reported earnings leap needs a quality check
| Consolidated K-IFRS | Q2 2025 | Q1 2026 | Q2 2026 |
|---|---|---|---|
| Revenue | 22.2320 | 52.5763 | 79.3187 |
| Operating profit | 9.2129 | 37.6103 | 60.5426 |
| Operating margin, calculated | 41.4% | 71.5% | 76.3% |
| Net profit | 6.9962 | 40.3459 | 93.9226 |
The earnings presentation separately identifies ₩63.27tn of investment-asset-related gains within non-operating items. That helps explain why net profit exceeds operating profit. We do not carry that gain into recurring earnings. Subtracting a pre-tax gain directly from after-tax net profit would also be an incomplete normalization because the associated tax and other adjustments need reconciliation. [2]
Our model starts with operating profit and applies explicit forward assumptions for non-operating income and tax. It does not annualize the headline net-profit figure. This preserves a clear distinction between manufacturing economics and financial-asset gains while recognizing that even operating margins currently reflect unusually favorable industry conditions.
Source limitation: we have not reconciled every historical EPS line in the summarized IR deck to attributable-profit and share-count footnotes in a complete reviewed interim filing. We therefore avoid a historical EPS-based valuation here. Our forward EPS proxy is calculated transparently from the assumptions below; it should not be confused with reported diluted EPS.
HBM4 and HBM4E should be tracked by commercial stage
In July, management said HBM4 mass shipments had begun in Q2, with a broader ramp planned for the second half. Separately, the June 18 release described 12-layer, 48GB HBM4E samples, with maximum 16Gbps-per-pin performance and more than 20% higher power efficiency versus previous models. Those are company specifications and sampling claims; they are not independent benchmarks or a disclosed HBM4E revenue forecast. [1] [3]
The September TSMC OIP report showed HBM4 and SOCAMM2 on a Vera Rubin exhibit and discussed joint packaging validation with TSMC. This supports the integration roadmap and partner relationship. An exhibition or validation milestone does not establish customer shipment volumes, exclusivity or the profitability of a production contract. [4]
The moat we would underwrite is repeatable execution across successive products. Design collaboration, thermal behavior, stack quality and usable output determine whether an impressive specification becomes a reliable customer supply program. Higher stack counts can add value while also increasing manufacturing complexity. We would look for stable yields and cash returns as the product mix changes.
Customer concentration belongs in this analysis even when the company has several customers. Demand can still be tied to a small number of accelerator roadmaps and hyperscaler budgets. The reviewed materials do not disclose enough customer-specific revenue to calculate a reliable current NVIDIA concentration percentage, so we assign none. A change in one large platform’s timing can matter without indicating a collapse in overall AI adoption.
Competition can expand the market and pressure returns together
Samsung announced commercial HBM4 shipments on February 12. Micron’s September 30 financial release describes increasing investment and strategic customer agreements, while projecting further growth in its fiscal 2027. These are separate company disclosures, not a market-share reconciliation. They show why SK hynix’s execution lead must be reassessed rather than assumed permanent. [5] [6]
Customers can benefit from multiple qualified suppliers through supply security and pricing negotiations. More suppliers can help a platform scale, expanding total demand, while reducing the incumbent’s share or negotiating power. The important outcome for shareholders is profitable revenue and cash flow, not market share viewed in isolation.
China-related competition can also affect conventional memory pricing without matching the newest HBM product immediately. Additional supply in one part of DRAM or NAND may change where established suppliers allocate capacity and how customers negotiate. We therefore avoid an all-or-nothing thesis in which advanced HBM leadership eliminates every source of memory cyclicality.
Cash generation is strong, but the definitions matter
| July IR cash-flow summary | Q1 2026 | Q2 2026 |
|---|---|---|
| Operating cash flow | 26.430 | 65.710 |
| Cash acquisition of PP&E | 7.657 | 10.671 |
| CFO less cash PP&E, AHNIM | 18.773 | 55.039 |
The large cash inflow is encouraging, but this limited subtraction is not a full measure of cash available for distribution. The presentation’s aggregate working-capital-and-other adjustment does not isolate every recurring and non-recurring component. We do not call the result normalized cash flow or equate it to the company’s shareholder-return definition of FCF.
June cash and short-term instruments were ₩87.958tn; the presentation’s defined borrowings were ₩18.587tn. The difference is ₩69.371tn, or approximately ₩69.4tn net cash under that definition. The cash balance itself is not net cash. The July equity raise and subsequent repurchases also mean June balances cannot represent the current balance sheet. [2]
Manufacturing growth consumes resources before it produces an accepted product. A supplier may need inventory, equipment, cleanroom capacity and technical staff ahead of customer collections. Receivables can rise during strong sales growth. We would assess cash conversion over multiple periods and examine deposits alongside delivery obligations, rather than treating any single quarter as a permanent run rate.
Capacity plans require a return on the money committed
The July presentation placed 2026 capex in the high-₩40tn range and discussed accelerating M15X and Yongin capacity. Its long-term agreements included pricing structures and financial mechanisms such as deposits. Those terms can improve visibility, but the disclosed summary does not establish that every future quantity or price is unconditional. [2]
The Indiana project is on a different timeline: the company’s August report targets locally packaged and tested supply from the second half of 2029. We give that plant no direct production contribution to our FY2027 earnings model. Investment spending and strategic benefits can occur well before its planned operating contribution. [9]
More capacity creates value when profitable demand absorbs its output. It can destroy value when multiple suppliers expand into the same demand forecast or when new processes require more investment than expected. We would compare capacity additions with customer commitments, utilization and incremental cash returns. An announced project budget is neither immediate revenue nor a completed production asset.
Solidigm and export policy add capital-allocation uncertainty
On October 1, SK hynix said Solidigm’s capital plans remained undecided and should be judged by their overall effect on existing shareholders. Its September 16 clarification similarly said reported U.S. memory-production arrangements with Intel had not been finalized. These statements are dated company positions, not a guarantee that discussions cannot progress. [8] [10]
Reuters reported on October 15 that Dalian equipment permissions were being linked to possible U.S. NAND investment. SK hynix’s response, quoted in that report, said equipment installation was authorized and no capacity expansion was being sought at Dalian Mod B. We preserve the difference between source-based reporting and the company’s response, and do not book a new factory or subsidiary IPO into our model. [11]
External subsidiary capital can finance growth while changing the parent’s share of future economics. Internal funding avoids one form of ownership dilution but uses cash that could serve another purpose. Neither structure is automatically superior. We would require actual terms, ownership, capital commitments and cash-transfer arrangements before assigning an incremental valuation benefit.
Shareholder returns and the share denominator
The August 19 announcement approved ₩40tn of repurchases for full cancellation and an ambition to return more than 50% of cumulative 2025–2027 FCF. It listed 730,492,365 issued ordinary shares and an approximately three-month repurchase window from August 20. Authorization is distinct from completed purchases and cancellation. [7]
Our valuation uses a 750 million economic-share budget, 2.7% above that disclosed gross issued count. This deliberately gives no advance EPS credit for the ongoing buyback and provides an allowance for equity-related claims. It is an AHNIM assumption, not a current reported diluted count, and it does not prove every potential future issuance is covered. The disclosed count is already after the July listing; we do not add the U.S. offering again.
The July 15 SEC filing confirms 17.79 million new ordinary shares, with a July 14 closing and actual issuance amount of approximately ₩39.89tn. These ordinary shares underpin the ADR issuance; they are not an additional independent share pool. [16]
Repurchases exchange cash for fewer claims on the business. We do not count both the entire cash balance and the full benefit of future cancellation in the same calculation. Our earnings-multiple model gives no separate cash-per-share uplift. Future evidence of actual cancellation can change the denominator, while its financing and interest effects need to be assessed together.
Our FY2027 assumptions and valuation formula
| AHNIM assumption | Bear | Base | Bull |
|---|---|---|---|
| Revenue, KRW tn | 350 | 480 | 580 |
| Operating margin | 40% | 55% | 65% |
| Operating profit, KRW tn | 140 | 264 | 377 |
| Net non-operating income, KRW tn | 4 | 4 | 4 |
| Effective tax rate | 25% | 25% | 25% |
| After-tax profit allocated to owners | 99% | 99% | 99% |
| Owners’ profit proxy, KRW tn | 106.920 | 198.990 | 282.8925 |
| Economic shares, millions | 750 | 750 | 750 |
| Economic EPS proxy, KRW | 142,560 | 265,320 | 377,190 |
| Applied P/E | 5× | 7× | 8× |
| Calculated value, KRW | 712,800 | 1,857,240 | 3,017,520 |
| Published value, KRW | 713,000 | 1,857,000 | 3,018,000 |
Formula: value = [(FY2027 revenue × operating margin + net non-operating income) × (1 − effective tax rate) × owners’ allocation] ÷ economic shares × P/E. Base: [(₩480tn × 55% + ₩4tn) × 75% × 99%] ÷ 750m × 7 = ₩1,857,240, rounded to ₩1,857,000.
Bear assumes substantial normalization in pricing and profit while AI demand still supports a large business. The 40% margin remains highly profitable, so this is not a severe industry-crisis scenario. A deeper downturn could produce losses beyond the Bear value. Its lower multiple reflects less confidence in earnings durability.
Base assumes demand and product mix support revenue expansion, but competition, supply response and reinvestment economics pull operating margins below the latest quarterly peak. It still requires exceptional profitability. The 55% margin is a valuation judgment under uncertainty, not a claim that management has guided a fall to that level.
The scale of normalization is material: Base FY2027 operating profit of ₩264tn is roughly flat against the accessed FY2026 consensus of ₩266.15tn, despite higher sales. This is a cautious earnings path within a growing business. A less severe margin normalization would support a higher value, as the sensitivity table shows. The FY2026 comparison is an external estimate, not a reported annual result.
Bull assumes stronger sales and more persistent pricing power, supported by successful product ramps and disciplined supply. Its higher multiple requires investors to gain confidence in repeatable earnings. A favorable revenue outcome accompanied by weak cash conversion would not satisfy that case on its own.
The ₩4tn non-operating input excludes extraordinary investment gains and is a simplified forecast of recurring net income outside operations. The 25% effective tax rate and 99% owners’ allocation are modeling assumptions, with a 1% allowance for minority interests; neither is asserted to equal the latest reported ratio. Our 5×–8× multiples reflect cyclicality and uncertainty rather than a measured peer average. We have not reverse-engineered inputs to match broker targets.
Consensus comparison: the earnings assumption explains most of the gap
The accessed Stock Analysis/S&P Global page reports a ₩3,141,025 one-year mean target from 39 analysts. It also displays FY2027 revenue of ₩531.45tn and adjusted EPS of approximately ₩470,320. Access date is October 15; individual broker dates vary and the page provides no single aggregate update date. The earnings forecast sample need not equal the target-price sample. [13]
Using the published Base, |₩1,857,000 − ₩3,141,025| ÷ ₩3,141,025 = 40.9%. Both target descriptions are broadly one-year horizons for the same Korean ordinary share, making this a useful screening comparison. The broker models may use different earnings years, accounting adjustments and valuation methods, so it is not a perfectly reconciled like-for-like model.
Our revenue is approximately 9.7% below the displayed forecast, while our EPS proxy is approximately 43.6% lower. Our 7× multiple is actually above the roughly 6.68× obtained by dividing the mean target by the displayed FY2027 EPS. That ratio is only a reference calculation: it does not show that every broker uses P/E. The gap therefore mainly reflects lower profit assumptions and differing earnings/share conventions, rather than a uniformly lower valuation multiple.
For a dated broker example, Mirae Asset’s original July 29 report maintained Buy and cut its target ₩4.2m → ₩2.8m, explaining a lower target P/B with unchanged earnings estimates. This is historical pre-results context from that report, not a new October downgrade. We did not establish a fresh October 15 target revision with an accessible original note. [14]
Sensitivity and a supporting cash-flow check
| One input changed from Base | Calculated value, rounded |
|---|---|
| P/E 6× | ₩1,592,000 |
| P/E 8× | ₩2,123,000 |
| Operating margin 50% | ₩1,691,000 |
| Operating margin 60% | ₩2,024,000 |
| Revenue 10% below Base | ₩1,674,000 |
| Additional 5% economic-share dilution | ₩1,769,000 |
An illustrative FY2027 unlevered cash-flow check is ₩264tn operating profit × 75% + ₩22tn depreciation/amortization − ₩65tn tangible/intangible cash investment − ₩30tn additional operating working capital = ₩125tn. The last three inputs are explicit AHNIM assumptions, not a verified facility-by-facility budget. Another ₩30tn of investment or working-capital absorption would reduce this result to ₩95tn.
This cash-flow exercise excludes net financing income and precedes minority-interest allocation. It tests whether the operating scenario could support investment; it is not distributable cash solely for ordinary shareholders. We have not built a complete multiyear DCF with a discount rate and terminal normalization, so it supplies neither a second price target nor independent confirmation of the earnings-multiple value.
What changes the case from here
TSMC’s October 15 release reported Q3 dollar revenue of $46.72bn and Q4 guidance of $52.3–54.3bn. This supports continued advanced-chip demand, but does not quantify SK hynix’s customer allocation, memory pricing or margins. SK hynix scheduled its own Q3 call for October 28 at 09:00 KST; those results had not been released at our cutoff. [12] [15]
Evidence for a higher case: product ramps that deliver attractive cash margins, durable customer commitments with clear economic terms, and revenue growth supported by collections. Completed cancellations could improve per-share economics, provided the cash cost and interest effect are included. A target increase should identify the changed input and evidence behind it.
Evidence for a lower case: conventional-memory price erosion, weaker HBM allocation or yields, deferred customer deployments, and expansion that absorbs more cash than expected. Export restrictions, subsidiary financing and equity compensation could reduce the common shareholder’s share of the operating upside. FX movements can also affect reported KRW results and competitiveness.
SK hynix offers meaningful exposure to a real AI-memory bottleneck. Our Base still requires a very profitable growth path, and the recent share-price fall does not remove the risk of a changing cycle. The central task for future updates is to measure how much of the current profit and cash generation survives through competition, capacity additions and capital returns.
Research history and related coverage
Related coverage: TSMC Q3 results: advanced-chip demand, 2nm margins and cash flow
Initial AHNIM SK hynix baseline: October 15, 2026. Future news updates will link here and state any changes to assumptions. SK hynix research history.
Samsung Electronics: AI memory and earnings durability
ASML: AI equipment demand and earnings quality
Astera Labs: AI connectivity and valuation. These examine different parts of the infrastructure chain and use company-specific assumptions.
This is general company research, not personalized investment advice. Scenario values depend on uncertain assumptions and can change materially. They do not limit potential investment losses.
Related coverage: Micron’s Taiwan HBM packaging opening: the production ramp, labor and cash-flow tests. This follow-up examines a separate part of the AI-memory supply chain; it does not revise the historical assumptions in this article.
Sources and original documents
[1] SK hynix — Q2 2026 results release, July 29, 2026; consolidated preliminary K-IFRShttps://news.skhynix.com/en/q2-2026-business-results/
[2] SK hynix — FY2026 Q2 earnings presentation, July 29, 2026; slides 6–7, 10–12, 15 and attachments; downloaded from official IR and visually reviewedhttps://mis-prod-koce-homepage-cdn-01-blob-ep.azureedge.net/web/attach/13571040283022005.pdf
[3] SK hynix — HBM4E sample announcement, June 18, 2026https://news.skhynix.com/en/sk-hynix-ships-samples-of-12-layer-next-gen-hbm4e-2/
[4] SK hynix — TSMC OIP portfolio and integration update, September 28, 2026https://news.skhynix.com/en/tsmc-oip-conference-2026/
[5] Samsung Electronics — HBM4 commercial-shipment announcement, February 12, 2026https://news.samsung.com/global/samsung-ships-industry-first-commercial-hbm4-with-ultimate-performance-for-ai-computing
[6] Micron — FY2026 results and FY2027 outlook, September 30, 2026; fiscal year ended September 3https://investors.micron.com/news/press-release/2026/Micron-Technology-Inc--Reports-Record-Fiscal-Fourth-Quarter-and-Full-Year-2026-Results/
[7] SK hynix — share repurchase/cancellation program and issued shares, August 19, 2026https://news.skhynix.com/en/share-buyback-and-retirement/
[8] SK hynix — Solidigm capital-plan clarification, October 1, 2026https://news.skhynix.com/en/fact-11/
[9] SK hynix — Indiana groundbreaking and planned supply timing, August 28, 2026https://news.skhynix.com/en/indiana-groundbreaking-ceremony-sketch/
[10] SK hynix — clarification on reported Intel discussions, September 16, 2026https://news.skhynix.com/en/fact-10/
[11] Reuters — Dalian equipment and potential U.S. NAND plans, October 15, 2026; includes SK hynix responsehttps://www.reuters.com/world/asia-pacific/sk-hynix-conditions-new-us-factory-more-us-chip-equipment-exports-china-sources-2026-10-15/
[12] TSMC — Q3 2026 results and Q4 guidance, October 15, 2026https://investor.tsmc.com/english/quarterly-results/2026/q3
[13] Stock Analysis / S&P Global — KRX 000660 close, analyst targets and forecasts, accessed October 15, 2026; per-broker dates vary and no single aggregate-update date displayedhttps://stockanalysis.com/quote/krx/000660/forecast/
[14] Mirae Asset Securities — Buy maintained, target reduced, July 29, 2026; original report, pages 1–2; historical contexthttps://securities.miraeasset.com/bbs/download/2146182.pdf?attachmentId=2146182
[15] SK hynix — Form 6-K filed October 14, 2026, Q3 earnings-call datehttps://www.sec.gov/Archives/edgar/data/2120882/000119312526420934/d413172d6k.htm
[16] SK hynix — Form 6-K, completed capital increase, July 15, 2026https://www.sec.gov/Archives/edgar/data/2120882/000119312526303972/d143606d6k.htm
Leave a Reply