Financial analysis of SK hynix, August 2026

SK hynix earns a 76% operating margin and trades at 8.8x trailing operating profit. The price requires about $64bn of free cash flow a year from 2029, roughly 3.5x what the company generated in 2025.

Built from the Form 424B4 IPO prospectus filed 9 July 2026, all 31 SEC filings to date, the 1H 2026 semi-annual business report and the Q2 2026 earnings call. Market data as of 28 August 2026. SK hynix reports in Korean Won under IFRS as issued by the IASB; every figure here is converted to US dollars on the basis set out below.

1. Currency basis

Two conventions are used, and each table says which. Historical statements (sections 2 to 5) are translated at the rates in force in each period: income and cash-flow items at the period average, balance-sheet items at the period-end noon buying rate, both as disclosed in the prospectus exchange-rate table. Valuation, per-share data, forward scenarios and announced future commitments (sections 6, 7, 9 and 11) are translated at the 28 August 2026 spot rate of ₩1,377.57, so that multiples are internally consistent and forecasts are not distorted by a currency view.

Federal Reserve Bank of New York noon buying rates as tabulated in the 424B4, plus the spot rate on 28 August 2026. Quarterly 2025 rates are not disclosed in the prospectus, so 2025 quarterly and half-year comparatives use the FY2025 average of ₩1,421.4. The prospectus uses a single convenience rate of ₩1,523.5 for all of its own dollar figures, so the FY2025 dollar amounts here differ from the ones printed in the filing; the Q1 2026 amounts match it exactly.

2. Financial performance, 2023 to 1H 2026

The audited record in the prospectus covers three years, which is the maximum a foreign private issuer must present on Form F-1. The 2023 loss year is the useful anchor: it shows what this business looks like at the bottom of a memory cycle, and it is only three years old.

Consolidated income statement, US$bn at period-average rates

Margins are currency neutral and equal the reported Won margins. Operating profit for 2023 to 2025 is computed as gross profit less SG&A less R&D, because IFRS as issued by the IASB does not present an operating-profit subtotal; the derived figures reconcile to the operating profit reported in the Korean semi-annual report. Q2 gross profit and cost of sales are derived by subtracting Q1 from the 1H figures. Net margin exceeds 100% in 2026 because of $42.6bn of gains on valuation and disposal of financial instruments, principally the Kioxia stake sold in June 2026, plus $9.4bn of dividend income. n.d. = not disclosed at that frequency.

Balance sheet, US$bn at period-end rates

The June 2026 period-end rate of ₩1,548.8 is 12.4% weaker than the 28 August spot rate, so the dollar balance sheet understates today's value of the Won-denominated position by roughly that margin. At spot, June equity is $190.7bn and net cash is $50.4bn; those are the figures used for valuation.

Cash flow and capital expenditure, US$bn at period-average rates

2026E capex of about $35bn is company guidance, given as "the high ₩40 trillion range" and converted at spot; it was raised on the Q2 call.

3. Ratio analysis

Percentages, multiples and days are currency neutral and match the Won-based figures. 1H26 annualised multiplies half-year flow items by two; balance-sheet ratios use the 30 June 2026 position. Return on equity uses average equity for 2024 and 2025 and closing equity elsewhere. Days inventory uses cost of sales. Per-share figures are converted at spot so that price, earnings and dividends share one basis. Each ADS is one tenth of a common share. Trailing 12 month earnings equal FY2025 less 1H2025 plus 1H2026, or $117.7bn at spot, on 728.9m shares outstanding.

What moved: bit volumes and prices, quarter on quarter

424B4 MD&A quarterly table through 1Q26; Q2 2026 earnings call and semi-annual report for 2Q26. Average selling prices in this table are disclosed by SK hynix on a US dollar basis, so they are unaffected by the translation convention. The full series back to 1Q 2023 appears in the prospectus.

4. Sector analysis

Company performance is not readable here without the industry frame. Memory is a three-player DRAM oligopoly and a five-player NAND market in which the marginal wafer is now allocated by a single customer group. The prospectus industry section, sourced to Gartner and IDC, is already denominated in US dollars.

Market size, US$bn, Gartner

2023 to 2025 DRAM and NAND figures come from the 1H26 semi-annual report (Gartner, June 2026). The 2026 and 2027 estimates come from the prospectus industry section (Gartner, 1Q26 update) and use a slightly different 2025 base of $216bn. Both are shown as filed rather than reconciled. Implied share divides SK hynix dollar revenue by the Gartner memory market for the same year and lands within two points of the IDC segment shares, which is a consistency check on the translation.

Competitive structure, Q1 2026, IDC revenue share

Peer benchmark, most recent reported quarter

SK hynix quarterly flows converted at the Q2 2026 average rate of ₩1,500.6; equity and price at spot. Micron figures from the 10-Q for the quarter ended 28 May 2026, accession 0000723125-26-000015; inventory days computed on that quarter's cost of sales. Samsung Electronics does not file with the SEC; its multiples come from Morningstar's comparison table and are not verified against primary filings. Morningstar's SK hynix multiples are computed on the ADR price, which trades at a premium to the Seoul line.

What is driving the shift

• Tiered AI memory. HBM next to the GPU, server DRAM next to the CPU, enterprise SSD underneath. Each tier scales with AI server count, so the three grow together rather than substituting. This is the mechanism behind enterprise memory going from 26.5% of the market in 2020 to a forecast 51.9% in 2027.
• HBM crowds out everything else. HBM consumes far more wafer per bit than conventional DRAM, plus through-silicon-via and packaging capacity. Producers reallocating cleanroom space to HBM created the conventional DRAM shortage. Gartner has traditional DRAM ASP up 45.2% year on year in Q4 2025, then 136.4% and 198.1% in Q1 and Q2 2026.
• NAND joined late and violently. Gartner expects NAND ASP up 111% year on year in Q1 2026, 244% in Q2 and over 250% in both Q3 and Q4. SK hynix NAND ASP rose mid-70% then mid-50% quarter on quarter in the last two quarters.
• HBM carries a real price premium. Gartner puts HBM above five times conventional DRAM per gigabyte in 2025. If conventional DRAM prices normalise faster than HBM, mix shift alone protects a large part of the margin.
• Barriers are capital, not just technology. 2026 industry capital spending of $237bn with 87.6% concentrated in twenty companies. New entrants cannot buy their way in at scale, and incumbents can only add supply on multi-year lead times, which is why the current shortage is slow to clear.

5. What management committed to on the Q2 2026 call

Where management pushed back

• On the AI capex slowdown narrative. Data-centre leasing and more efficient models are framed as monetisation and utilisation of infrastructure already built, not reduced investment. Management said major customers are still requesting more memory supply.
• On oversupply from its own expansion. Capacity is planned against demand visibility secured in customer partnerships, and equipment investment and ramp will be phased. Management does not expect its medium-term plan to cause oversupply "right away." The qualifier leaves 2028 open.
• On HBM competition. The claimed differentiator is not performance alone but yield, quality and the ability to supply volume on time, described as not replicable quickly.
• On the weak DRAM blended ASP in Q2. Attributed to mix, with some high-value shipments slipping into the second half. Management expects both higher bit growth and better mix in H2, so blended ASP should improve.

Capital commitments approved since the IPO, US$bn at spot

Board-approved future spending is converted at spot because it will be incurred over 2026 to 2032. The 30 June commitments balance is converted at the June period-end rate to stay consistent with the balance sheet; at spot it is $44.6bn. The 50.9% ratio is the company's own disclosure against FY2025 equity and is currency neutral. The long-range plan announced on 29 June 2026 indicates roughly $436bn for the Yongin cluster, $290bn for a new southwestern cluster and $73bn for Cheongju, around $800bn in total over multiple decades. Those are announced intentions, not contracted commitments.

6. Scenario analysis, 2026 to 2028

Revenue is modelled bottom up as DRAM and NAND bit volume multiplied by average selling price, because that is the form in which the company guides and discloses, and because its ASP series is already on a US dollar basis. Cost of sales is split into a cash component that scales with bits and unit cost, and depreciation driven by the capex schedule. A profit-sharing charge of 9% of pre-bonus operating profit is applied, reflecting the incentive programme the MD&A identifies as the main driver of rising cost of sales. Tax is held at 24%, close to the 23.0% effective rate in 1H 2026. All forecast figures are translated at ₩1,377.57 and assume that rate holds.

Core assumptions by scenario

Cash cost per bit rises in the bull and base cases because HBM4 and HBM4E are more wafer- and packaging-intensive per bit than conventional DRAM, and because the mix is shifting toward them. It falls in the bear case on aggressive node migration and cost programmes. Non-operating income is set at $48.6bn in 2026, being the realised Kioxia and financial-instrument gains booked in 1H26 plus a small H2 contribution, and normalises to $1.8bn thereafter, $0.4bn in the bear case.

Scenario outputs, US$bn unless stated

Shares fall from 728.9m to 704.8m as the $29bn buyback completes and is cancelled by November 2026; 2026 EPS uses a part-year weighted count. 2026 net income includes the one-off financial gains; excluding them, base-case 2026 core EPS is $197.13 per common share, or $19.71 per ADS, a P/E of 6.1x. Free cash flow is EBITDA less cash tax, capex, and a 5% of revenue allowance for working-capital build, which is conservative given receivables have grown faster than sales. Divide per-share figures by ten for the per-ADS equivalent. The last row divides today's price by an undiscounted future-year figure, so it states what a buyer at today's price would have paid per dollar of that year's earnings if the scenario is realised. It is not a valuation; section 11 discounts the cash flows.

Assumption leverage, ranked

• DRAM ASP in 2027 dominates the model. Moving 2027 DRAM ASP from +5% to -25% removes about $84bn of operating profit, roughly a third. Nothing else in the model has that leverage.
• Bit growth matters far less. A ten-point swing in 2027 DRAM bit growth, from 25% to 35%, moves operating profit by about $20bn, a quarter of the ASP effect. When gross margin is 77%, incremental bits are almost pure profit but incremental price is total profit.
• Depreciation is the delayed bill. D&A rises from $9.8bn in 2025 to $26bn by 2028 in the base case, a $16bn annual fixed-cost step arriving on a two-year lag from today's capex decisions.
• Currency is third order. A 10% dollar move is roughly $1.9bn of pre-tax profit, about 1% of base-case 2027 operating profit. It matters more for reported Won earnings than for the dollar investment case.

7. Sensitivity analysis

The grid holds the base-case 2026 exit cost structure and varies only 2027 DRAM ASP and DRAM bit growth. It shows the price paid per dollar of 2027 earnings under each outcome, not a discounted value. NAND moves sympathetically, with bit growth at 0.8x the DRAM rate and ASP at 0.9x the DRAM change. ASP changes are on the company's own US dollar basis, so the grid is unaffected by the translation convention.

Other sensitivities disclosed in the filings

As of 31 March 2026, per the MD&A market-risk disclosure, converted at that date's rate of ₩1,523.5. The dollar has since fallen 10.4% against the Won, from ₩1,538 on 2 July 2026 to ₩1,378 on 28 August 2026. On the disclosed sensitivity that is roughly a $1.9bn annual headwind to pre-tax profit, and it postdates the 29 July guidance.

8. Critical factors, ranked by how much they move the bull case

9. The ADR anomaly

SK hynix listed 17,790,000 new shares as American Depositary Shares on Nasdaq on 10 July 2026 at $149.00 per ADS, raising $26.507bn gross and roughly $26.2bn net. Management described it on the Q2 call as the largest US IPO ever by a foreign company. Each ADS represents one tenth of a common share.

The mechanism is disclosed on the Q2 call. ADRs convert freely into Korean common shares, but converting shares into ADRs requires a regulatory filing that can take several weeks and is capped at 17,790,000 shares, the exact size of the offering, which is already fully used. The arbitrage that would close a premium, shorting the ADR and buying Seoul stock to deposit into the programme, is blocked. Management said no decision has been made on raising the ADR ratio.

10. Risks

11. Valuation

Today's price and a future year's earnings are different objects. Dividing one by the other measures an entry price, not a value. Two separate questions follow, and they have different answers.

What you pay, at ₩1,377.57

Question one: is it cheap on realised performance alone?

The answer inverts depending on the base. On the last twelve months the stock is on 7x to 12x. On any multi-year average of realised operating profit it is on 24x to 54x, because three years ago the company lost $5.9bn at the operating line. A trailing multiple on a cyclical at peak earnings locates the position in the cycle. It does not price the asset.

Last twelve months means July 2025 to June 2026, computed as FY2025 less 1H2025 plus 1H2026, converted at the 28 August spot rate so that price and earnings share one basis. Averages use each year's operating profit converted at the same rate, which removes currency from the comparison.

Question two: is it cheap on anticipated performance?

This one requires the future discounted back. Taking the modelled base-case free cash flow for 2027 and 2028, discounting at 10.5% and treating everything from 2029 as a perpetuity growing at 2%, the two forecast years account for $229bn of the $826bn enterprise value. The remaining $597bn is the terminal value, and it implies a normalised free cash flow of $64bn a year from 2029.

Run the other way, holding the base path, a 10.5% discount rate and 2% growth:

The 2027 and 2028 estimates move the answer very little, since together they are 28% of enterprise value. What decides it is whether AI memory demand holds free cash flow above roughly $64bn a year once the current price spike ends. On the 2025 base of $18.2bn the shares are worth about half of today's price. At the modelled 2028 peak of $125bn they are worth 65% more.

The 10.5% discount rate is an assumption, built from a Korean risk-free rate, an equity risk premium and a beta above one. None of those three inputs has been verified against a source here, which is why the table spans 8.5% to 12.5%. Cash flows are discounted from 29 August 2026 with 2027 and 2028 treated as arriving mid-year and the terminal value struck at the end of 2028. Nothing beyond 2028 appears in any filing; the perpetuity is a modelling device, not a forecast.

Three numbers

What would change this view

• Positive. Disclosure of LTA coverage as a percentage of sales, at any level above roughly a third. That single number converts a spot-priced commodity into a contracted revenue base and narrows the range of plausible normalised free cash flow.
• Positive. Q3 2026 results confirming both the 10% DRAM bit growth guide and improved blended ASP from HBM4 mix, validating the second-half ASP recovery management promised.
• Negative. Failure to obtain the 2027 annual BIS licence for the China fabs. Wuxi and Dalian are not marginal assets.
• Negative. Evidence that HBM4 yields are improving fast enough to release conventional DRAM wafer capacity in 2027, which would unwind the shortage that is generating most of the profit.
• Negative. Capex guidance for 2027 materially above $50bn without a corresponding LTA disclosure.
• Currency. Further Won strength compresses dollar earnings roughly one for one on translation and by about $1.9bn of pre-tax profit for each 10% move on the transaction exposure. A weaker Won does the reverse.

12. Sources and method

Primary sources

• SK hynix Inc., Form 424B4, filed 9 July 2026, accession 0001193125-26-299963. Exchange rates used here come from its Exchange Rates section.
• SK hynix Inc., Forms 6-K, sixteen filings between 15 July and 21 August 2026, including the Q2 preliminary results of 29 July, the semi-annual business report of 18 August and the buyback, cancellation and shareholder-return filings of 19 August.
• SK hynix Q2 FY2026 earnings conference call transcript, 28 July 2026; original recording SK hynix Investor Relations.
• Micron Technology Inc., Form 10-Q for the quarter ended 28 May 2026, accession 0000723125-26-000015. XBRL financial statements, reported in US dollars.
• Market data: KRX close for 000660 and USD/KRW spot, 28 August 2026; Nasdaq close for SKHY and Micron, 28 August 2026. Samsung Electronics multiples from Morningstar comparison data, not verified against primary filings.
• Third-party industry data as cited within the prospectus and semi-annual report: Gartner and IDC, already denominated in US dollars.

Method and limitations

• All company financials are as filed with the SEC in Korean Won, then translated as described in section 1. Percentages, margins, ratios and multiples are currency neutral and match the Won-based figures.
• Operating profit for 2023 to 2025 is derived as gross profit less SG&A less R&D, because IFRS as issued by the IASB does not require an operating-profit subtotal. The derived figures reconcile to the operating profit reported in the Korean semi-annual report for 2024 and 2025.
• Q2 2026 gross profit and cost of sales are derived by subtracting the Q1 2026 interim figures from the 1H 2026 semi-annual figures. Q2 2026 results are provisional; the 6-K states the external auditor's review was not complete.
• Quarterly 2025 exchange rates are not disclosed in the prospectus, so 2025 quarterly and half-year dollar comparatives use the FY2025 average rate.
• Segment revenue is not disclosed quarterly after Q1 2026. The DRAM and NAND split used in the model from Q2 2026 onward is derived from disclosed bit-volume and dollar ASP changes, calibrated so that the modelled Q2 total equals the reported figure exactly.
• Only three years of audited annual financials exist in the public record, because Form F-1 requires three years for a foreign private issuer. A five-year series is not available from SEC sources.
• Scenario outputs are the author's model, not company guidance, and hold USD/KRW flat. Where the company has guided, the guidance is used directly and identified as such.
• The valuation in section 11 discounts modelled 2027 and 2028 free cash flow from 29 August 2026, treating each year as arriving mid-year, and strikes a perpetuity at the end of 2028. The 10.5% discount rate and the 2% perpetuity growth rate are assumptions; neither the risk-free rate, the equity risk premium nor the beta behind them has been verified against a source, which is why the tables span 8.5% to 12.5%.
• Price divided by a future year's earnings, shown in sections 6 and 7, measures the entry price against one modelled year. It is undiscounted and is not a valuation.
• This is analysis of public filings. It is not investment advice.