SKHY · Forward model · Song case

The Song case, 20 quarters out

Each segment is projected from its own operating driver, rolled up into consolidated cash flow, and discounted back to a fair value per share. The assumptions below are editable — change them and every number on this page moves with them.

Quarters are calendar quarters, the way SK hynix reports them: 2026 Q2 is the three months ended 30 June 2026, reported 29 July 2026. SK hynix reports ONE segment - 'as revenue from the semiconductor segment exceeds 90% of the Company's total revenue, its business is presented as a single semiconductor segment' - but every quarterly, half-year and annual filing carries a revenue-by-product note splitting revenue into DRAM, NAND Flash and Others. Those three lines are this model, and nothing finer. HBM is 56.4% of a market SK hynix leads and it appears in no revenue table in any filing, so there is no HBM vertical here; inventing one would be manufacturing the most important number on the page. Fourteen quarters of history, 2023 Q1 to 2026 Q2, sum to reported consolidated revenue in every quarter, and to $52.06bn (W79,318,746m) in the basis quarter. Eleven of the fourteen are read from a disclosed three-month column; the three December quarters are derived as full year minus nine months and are the only points marked estimated - each still ties to the quarterly release total (W11.31tn for 4Q23, W19.77tn for 4Q24, W32.83tn for 4Q25). Currency: the company files in won and this model runs in USD at a FIXED W1,523.5, the rate the stored series was built on, so quarters stay constant-currency. The market price it is measured against embeds a spot rate near W1,382, about 10% stronger, and the ADS at $158.02 sits about 26.5% above the Seoul common line translated at that spot - both gaps are real and neither is modelled away. Volume is an INDEX, not a bit count: SK hynix discloses only quarter-on-quarter percentage changes in shipments, never absolute bits, so units are set to 100 at 2026 Q2 and the price per index point is that quarter's revenue divided by 100. Read the unit traces as relative volume and relative ASP. No product-level margin, capex or capacity is disclosed anywhere, so the 81.4% basis EBITDA margin is the consolidated figure (operating profit W60,542,608m plus D&A W4,028,246m over revenue) applied to both memory lines, and the terminal margins - 45% for DRAM, 30% for NAND, 15% for Others - are assumptions about a mid-cycle, not disclosures. Because that margin is already after SG&A, corporate overhead is set to zero rather than charged twice. Capital spending is split: 9.5% of revenue at the basis quarter gliding to 15% carries the maintenance and technology-migration spend inside the verticals, and the two fab programmes approved on 7 August 2026 - Yongin Y2 at W35,224.6bn to October 2031 and Cheongju M17 at W19,100bn to April 2031 - are carried separately at their filed totals and schedules, because a committed fab spends on its own calendar whether or not revenue grows. Together they put FY2026 capex at about W46 trillion against guidance of the high W40 trillion range. P&T7's W7,093.1bn is an increase to an already-approved packaging plant with no disclosed schedule, so it sits inside the revenue-scaled line rather than as a third programme. Share count and net cash are both pro-forma for the same November 2026 state, so the buyback is not counted twice: 7,047,966,690 ADS-equivalent shares (728,866,669 common less the 24,070,000 approved for cancellation, times ten ADSs per share) and $45.3bn of net cash (W69.4tn at 30 June, plus the W39.9tn of July ADS proceeds, less the W40.0tn buyback and the W0.27tn Q2 dividend). Both assume the programme completes in full by 19 November 2026. Net profit is NOT in this model at all: the W93.9tn reported for the quarter exceeds revenue because W65.9tn of it is finance income, largely a derivative revaluation on the 2023 exchangeable bonds. Seasonality: DRAM and NAND are modelled aseasonal on purpose. A ratio-to-centred-moving-average run over all fourteen quarters gives DRAM factors of [0.927, 1.051, 1.014, 1.008] with window-to-window spreads of 0.051, 0.059, 0.191 and 0.252 - the spread exceeds the signal by two to five times, and the apparent Q4 strength is trend, because every year in the sample was an up-year. Only Others carries a shape bigger than its noise, and it is half a percent of revenue.

Where each case comes from

From cash flow to fair value

Capex outside the verticals

Each vertical's capex is a share of its own revenue, which is the right shape for capacity that scales with sales. These are not that: fixed-size programmes with their own schedule that spend whether or not any segment grows. They are added on top, and they are why free cash flow dips in the years below.

The projected path

Every row is projected. A year-over-year change is shown only where the quarter it compares against exists — an em dash means there is no comparable quarter, not a flat year.

Model revisions

Assumptions are marked to reality as each quarter prints. Every change is appended here, with the fair value the model produced at the time, so the model's own history stays visible.