Memory semiconductors. SKHY (Nasdaq ADS; KRX 000660) / SNDK (Nasdaq). Run 11 September 2026. Built from the SK hynix 424B4 prospectus (July 2026), semi-annual report 6-K (18 Aug 2026) and Q2 2026 call; the Sandisk FY2025 and FY2026 10-Ks, Q4 FY2026 results 8-K, Q4 call, Investor Day (13 Aug 2026) and September 2026 conference transcripts; plus cited TrendForce, Counterpoint and press sources. Events swept through 11 Sep 2026; most recent events checked: Kioxia publicly dismissing an SK hynix tie-up (Bloomberg, 9 Sep 2026) and Sandisk at Goldman Sachs Communacopia (9 Sep 2026). Absolute figures in US$; SK hynix won converted at W1,523.5/US$ (424B4 convenience rate, 31 Mar 2026); ratios are FX-free. Not a valuation and not a recommendation.
Three SK hynix cells carry the next five years.
New wafers feed all three: M15X has been pulled in and Yongin Y1 opens in February 2027 (TrendForce, 14 Jul 2026). Sandisk's forward engine is narrower. Its FY27 bit growth of mid-to-high teens is "100% driven by node transition", with no wafer additions (Citi, 8 Sep 2026). Its largest cell, Edge NAND shipped to Asia, is growing on price while smartphone and PC units fall mid-teens (Q4 FY26 call). The past growth rates in the chart below are evidence, not the reason.
Through the cycle, SK hynix converts better, for two reasons.
The tension sits in the latest matched quarter, April to June 2026. Sandisk printed 84.6% gross and 78.5% operating margin, against SK hynix's 83.2% and 76.3%. Two things explain it. NAND prices rose faster than DRAM, and SK hynix's HBM is priced in annual contracts that lag spot (Q2 2026 call). Sandisk's capital-light JV also turns more revenue into cash at the peak: FY26 free cash flow was 43% of revenue excluding prepayments (Q4 call), against 27% for SK hynix in CY2025 (424B4). If Sandisk's NBM floors hold ("around 80%" gross margin even at floor pricing), its trough could sit structurally higher than its history suggests. This is the most important open question in the duel.
SK hynix's exposure: fixed cost meeting a thinner HBM premium. Its HBM share fell from 58% to 50% in one quarter as Samsung rose from 21% to 33% (Counterpoint via JoongAng Daily, 3 Sep 2026). Meanwhile about W104tn of Korean fab and packaging projects land between 2027 and 2031: Y1 and P&T7 at W50tn, Y2 and M17 at W54tn (424B4; SK hynix newsroom, Aug 2026). The China fabs need a US licence every year.
Sandisk's exposure: the unprotected half of its book. The contracted NBM bits (a little over half in FY27) are protected. The rest is mostly Edge and Consumer, 75% of FY26 revenue, priced off spot, in device markets where memory inflation is already killing low-end models (Citi). It still owes half the JV's fixed costs whatever it buys. In FY2023 its gross margin was 7.1% (FY25 10-K).
Why Sandisk breaks first. Its spot exposure reprices within a quarter, while SK hynix's HBM reprices annually and is backed by W69.4tn of net cash (Q2 call). SK hynix breaks harder only if it keeps building into a downturn.
These five cells carry over 80% of the combined economics. Cells pair sourced product and region splits; neither company discloses the cross-tab, so the pairing is inferred and the logic is given in the Growth Map tab.
| Cell (product x region) | SK hynix | Sandisk | Why (one clause, sourced) |
|---|---|---|---|
| AI memory (HBM + server DRAM) x US accelerator and cloud buyers | 4 | 0 | SK hynix is #1 in HBM at 50% but ceded 8 pts in 2Q26 (Counterpoint). Sandisk's HBF only samples in 2027 (Investor Day). |
| Datacenter NAND (eSSD) x US hyperscalers | 3 | 3 | SK hynix is #2 at 23.0%, 3x Sandisk's revenue. Sandisk is #5 at 7.9%. Both grew roughly with the market in 2Q26 (TrendForce, 1 Sep 2026). |
| Client and mobile NAND (Edge) x China, HK and other Asia | 3 | 3 | This is Sandisk's largest cell (Edge is 60% of revenue; Asia is 70%), growing on price while units shrink. SK hynix's China NAND pricing is capped by SAMR through Dec 2026 (424B4). |
| Conventional DRAM x China and Asia OEMs | 3 | 0 | SK hynix's DRAM share went from 34.8% to 29.1% as capacity moved to HBM (IDC via 6-K 18 Aug 2026). Sandisk has no DRAM. |
| Retail flash x global | ND | 4 | Sandisk has 351,000 points of sale and gained 2 pts of consumer share since Feb 2025 (Investor Day). SK hynix discloses no retail line. |
Every score is anchored to an exhibit in the tabs below. SK hynix's HBM gets a 4 rather than a 5 because its share fell in 2Q26, and rather than a 3 because it is still the default supplier at #1. Scores are per cell and are never summed.
Normalization first. SK hynix splits revenue into DRAM, NAND and other, and books region by the location of its sales entity. Sandisk splits revenue into Datacenter, Edge and Consumer (all NAND), and books region by ship-to address. So SK hynix's "US" includes accelerator sales shipped to Asian assembly sites, and Sandisk's "Hong Kong" is largely a transit point for China and Asia (inferred). No cell-level cross-tab exists for either company. The cells below pair sourced splits and are marked inferred.
The butterfly shows the contrast in one picture. Almost 69% of SK hynix revenue is booked in the US. About 70% of Sandisk's ships to China, Hong Kong and other Asia, and its fastest growth came from there on Edge pricing (FY26 10-K: "higher revenue in the Asia and Americas regions from Edge and Datacenter customers, respectively").
Product mix, latest period
Top bar: SK hynix Q1 2026 (424B4). Bottom bar: Sandisk FY2026 (10-K). SK hynix does not split NAND into eSSD and client (ND).
| Cell | SK hynix revenue | SK hynix growth | Sandisk revenue | Sandisk growth | Margin signal / leader |
|---|---|---|---|---|---|
| AI memory x US | Largest cell (inferred: DRAM 77%, US 68.8%, largest customer 23.9% of 2025) | DRAM +90% CAGR 2023-25 | 0 | n/a | SK hynix. HBM is priced at more than 5x conventional DRAM per GB (Gartner via 424B4). |
| Edge NAND x Asia | ND | ND | Edge US$12.16bn, 60% (10-K) | +195% FY26; exabytes +high single digits | Price-led: revenue per GB up about 180% (10-K). |
| Datacenter NAND x US | eSSD >US$8.63bn in 2Q26 (TrendForce) | +86% QoQ (derived) | US$5.15bn FY26, 25%; about US$2.98bn in 2Q26 | +437% FY26 | SK hynix on scale. Sandisk gaining mix: datacenter went from 12% to 38% of bits (Q4 call). |
| Conventional DRAM x China/Asia | China 19.7% + Asia 7.4% of total | China +23% CY25 | 0 | n/a | SK hynix vs Samsung and Micron |
| Retail flash x global | ND | ND | US$2.94bn, 14.5% | +29% FY26; Q4 -32% QoQ | Sandisk on brand |
| All other | Foundry/other 1.6% | -29% CY25 | none |
HBM (SK hynix only). Growth comes from content per accelerator and generational steps. NVIDIA is reported to allocate about two-thirds of Rubin HBM4 to SK hynix (TrendForce, 28 Jan 2026; a report, not a disclosure). Growing here takes packaging capacity: P&T7 at W19tn and a US$4bn Indiana HBM plant producing from 2029 (424B4; TrendForce, 28 Aug 2026). The growth is organic.
Datacenter NAND (both). SK hynix's eSSD franchise was bought (Intel NAND, US$8.8bn) and has since grown: Solidigm's 30TB+ QLC revenue tripled in Q2 (Q2 call). Sandisk's is organic. Its Stargate QLC platform shipped for revenue in Q4, and its TLC and QLC drives are qualified at major hyperscalers (Q4 call; Investor Day).
Edge NAND (Sandisk). The driver has flipped from volume to price. FY26 Edge exabytes grew high single digits while revenue per GB rose about 180% (10-K). Management expects unit growth to return in CY2027 (Q4 call).
SK hynix sells direct through its sales subsidiaries and has cut its distributor reliance (424B4). It has about ten LTAs of roughly five years with deposits, but will not say what share of sales they cover (Q2 call).
Sandisk sells to OEMs, cloud providers, distributors and retailers. Price protection and incentives took 11% of gross revenue in FY26, down from 19% in FY25 and FY24 (10-K). Its eight NBMs cover more than 50% of FY27 bits and about two-thirds of FY28. They mix fixed and variable pricing with floors and ceilings, and management expects "around 80%" gross margin at floor. Minimum TCV is US$93.9bn, backed by US$16.5bn of guarantees (Q4 call).
| Item | SK hynix | Sandisk |
|---|---|---|
| Wafer source | Own fabs in Korea (Icheon, Cheongju) and China (Wuxi DRAM, Dalian NAND) (424B4) | 100% from the Flash Ventures JV at cost plus a small markup; owes half the fixed costs regardless of output (10-K) |
| Single points | China fabs (about 30-35% of DRAM and 35-40% of NAND capacity, TrendForce est.) run on an annual US licence | One partner, with JV terms extended to 2034. Kioxia's largest holder is an SK hynix vehicle at 14.19% (Korea Herald, 12 Aug 2026) |
| Input exposure | Sells DRAM | Buys DRAM for SSDs; took a 4% Nanya stake for access (Investor Day) |
| Capex / revenue | 28.3% (CY25, 424B4) | 2.2% cash (FY26 10-K); about 6% including its JV share (Q4 call) |
HBM manufacturing. First to mass-produce HBM3 and HBM3E; HBM4 is at yields near HBM3E (Q2 call); share was 50-58% in 2026 (Counterpoint). Durability: Medium-High. A rival needs years of qualification, but Samsung went from 21% to 33% in one quarter.
Two-product scale with owned fabs. #2 in DRAM (29.1%) and NAND (18.5%) (IDC via 424B4), with W69.4tn of net cash (Q2 call). Durability: High.
JV cost scale. Together with Kioxia it produces "roughly a third of the world's NAND" (GS, 9 Sep 2026). It claims 29% of industry bits on 13% of industry capital over 2021-25 (Investor Day; company claim, approx., unverified). Durability: Medium, because the scale is shared with Kioxia.
Consumer brand. 351,000 points of sale and +2 pts of share (Investor Day). Durability: Medium-High, but the cell is small (14.5% of revenue).
SK hynix is concentrated. Its largest customer was 23.9% of 2025 revenue, and its top two were 14.8% and 12.4% in Q1 2026 (424B4).
Sandisk is diversified and contracted. No customer passed 10% in FY24-FY26, and the top ten were 44% (10-K). Its eight NBM customers include three US hyperscalers; guarantees total US$16.5bn, of which US$2.9bn is cash deposits and credits. Q4 operating cash flow included US$1.94bn of prepayments (Q4 call; Investor Day).
On disclosed terms, Sandisk has the better customer book. SK hynix has the better product.
SK hynix owns its supply. It can move wafers between DRAM, HBM and NAND. Its exposure is the annual US licence for its China fabs; the 2026 licence was granted (Tom's Hardware, 30 Dec 2025).
Sandisk shares its supply. It buys every wafer from Flash Ventures, owes half the fixed costs whatever it takes, and is barred from third-party NAND manufacturing (10-K). It also buys DRAM for its SSDs from the companies it competes with. An SK hynix vehicle is reported as Kioxia's largest holder (Korea Herald, 12 Aug 2026), and Kioxia has publicly dismissed a tie-up (Bloomberg, 9 Sep 2026).
Neither company passed costs through in the last downturn: SK hynix's CY2023 gross margin was -1.6% and Sandisk's FY2023 was 7.1%.
| Cell | Route control | Pocket price | Continuity | Outcome (share / margin) | Confirming KPI |
|---|---|---|---|---|---|
| AI memory x US | SK hynix direct; Sandisk absent | SK hynix at a premium (more than 5x DRAM per GB) | SK hynix: own fabs, Korean packaging | SK hynix share down (58% to 50%), margin flat to up | HBM share at or above 45% by Q2 2027 |
| Datacenter NAND x US | Both direct: Sandisk via NBMs, SK hynix via LTAs and Solidigm | Sandisk floor-protected; SK hynix undisclosed | SK hynix own fabs; Sandisk JV | Sandisk share and margin up from a low base; SK hynix share flat to down (25.1% to 23.0%), margin up | TrendForce eSSD share, Sandisk above 10% by 2Q27 |
| Edge NAND x Asia | Sandisk: OEMs, distributors, and the Unis venture in China | Sandisk at spot less 11% incentives; SK hynix capped in China by SAMR | Both exposed to China policy | Sandisk share flat; its margin is the first to fall in a turn | Sandisk Edge revenue per GB; incentive % |
The tension. Sandisk grows fastest in datacenter, where it is smallest (#5, 7.9%). Its largest cell, Edge, is the one with the least price protection.
Mapping. Both companies report costs by function, so no by-nature proxy is needed. SK hynix's operating profit is gross profit less selling and administrative and R&D; in the 6-K, R&D sits inside S&A and was split out from Note 22. For Sandisk, EBIT is gross profit less R&D and SG&A, which excludes itemized one-offs (a US$1.83bn goodwill impairment in FY25, separation costs and other items). GAAP operating margin was -7.0% in FY24, -18.7% in FY25 and 61.3% in FY26.
Windows. SK hynix is measured over CY2024, CY2025 and LTM June 2026; Sandisk over FY2024 to FY2026, with years ending in June. The first two pairs are offset by six months and the third is matched.
| % of sales, 3y avg | SK hynix | Sandisk | Gap | What drives it |
|---|---|---|---|---|
| COGS | 38.4 | 60.8 | -22.4 | DRAM/HBM mix, and SK hynix's upcycle started a year earlier |
| R&D | 6.1 | 12.6 | -6.5 | Scale: a full NAND roadmap funded on one-sixth of the revenue |
| SG&A | 4.8 | 6.0 | -1.2 | Sandisk's retail channel and brand spend |
| Gross margin | 61.6 | 39.2 | +22.4 | Scarcity pricing in HBM; NAND repriced later |
| EBIT margin | 50.7 | 20.6 | +30.1 | Gross margin gap plus operating leverage on scale |
| Year | COGS | R&D | SG&A | Gross margin | EBIT margin |
|---|---|---|---|---|---|
| SK hynix CY2024 | 51.9 | 6.7 | 5.9 | 48.1 | 35.5 |
| SK hynix CY2025 | 39.6 | 6.7 | 5.2 | 60.4 | 48.6 |
| SK hynix LTM Jun-26 | 23.7 | 4.9 | 3.3 | 76.3 | 68.0 |
| Sandisk FY2024 | 83.9 | 15.9 | 6.8 | 16.1 | -6.7 |
| Sandisk FY2025 | 69.9 | 15.4 | 7.8 | 30.1 | 6.9 |
| Sandisk FY2026 | 28.5 | 6.6 | 3.3 | 71.5 | 61.6 |
Trough reference, one year earlier: SK hynix CY2023 gross margin -1.6% and EBIT -23.6%; Sandisk FY2023 gross margin 7.1% and EBIT -21.3%. Capex as a share of revenue: SK hynix 25.4%, 24.1% and 28.3% over CY2023-25 (424B4); Sandisk 2.2% in cash in FY26, about 6% including its JV share (10-K; Q4 call).
Gross margin is the persistent gap. SK hynix led by 32.0, 30.3 and 4.8 points across the three windows. Mix explains it: DRAM and HBM carry scarcity pricing, and NAND is all Sandisk sells. The gap has almost closed at the peak because NAND prices rose faster (NAND ASP up mid-70% then mid-50% quarter on quarter, against DRAM's mid-60% then about 30%; SK hynix Q2 call) and HBM contracts lag spot.
The cross-check supports the power map through the cycle. The latest quarter seems to contradict it, but that is contract timing, not Sandisk pricing power.
Capex points the other way. SK hynix carries its own depreciation (W13.9tn in CY2025, 14.3% of revenue; 424B4). Sandisk's JV depreciation reaches it through wafer cost, so the peak shows up in its free cash flow: US$8.7bn, 43% of FY26 revenue excluding prepayments (Q4 call).
| Metric | Threshold | By when | If it hits, it favors | Where published |
|---|---|---|---|---|
| SK hynix HBM revenue share | At or above 45% (flip below 40%) | Q4 2026 and Q2 2027 prints | SK hynix | Counterpoint / TrendForce quarterly |
| Sandisk gross margin in a falling-price quarter | At or above 70% non-GAAP | Through FY2028 (Jun 2028) | Sandisk | Sandisk 8-K results |
| SK hynix 2027 capex vs DRAM ASP | Capex guided above 2026 while DRAM ASP falls quarter on quarter | Q4 2026 / Q1 2027 calls (Jan-Apr 2027) | Sandisk (SK hynix weakens) | SK hynix calls |
| Sandisk datacenter mix and eSSD share | Datacenter at or above 40% of FY27 revenue; eSSD above 10% of top five | FY27 10-K (Aug 2027); TrendForce 2Q27 | Sandisk | 10-K; TrendForce |
| US licence for SK hynix China fabs | 2027 licence granted, unchanged in scope | 31 Dec 2026 | SK hynix | SK hynix 6-K; US Commerce |