No probability history is available for this market yet.
SK Hynix annual revenue
Calculation of starting probabilities
The starting probabilities of this market were calculated by combining SK hynix's own revenue history, a memory-chip peer cycle, and near-term AI-memory demand signals. The market resolves in Korean won because SK hynix's official K-IFRS releases report consolidated revenue in KRW; converting to USD would add exchange-rate noise that is not part of the question.
Latest completed fiscal-year anchor: SK hynix reported KRW 97.1tn of FY2025 revenue, up 47% year over year from KRW 66.2tn in FY2024. Its 4Q25 revenue was KRW 32.8tn, equal to a simple annualized run rate of about KRW 131.3tn. The company also said HBM revenue more than doubled year over year and that it had begun large-scale HBM4 production to meet customer requests.
Key charts


Reference classes used
- SK hynix's own cycle: 10 year-to-year observations from 2015-2025. This captures the company's direct exposure to memory booms and downturns, including the 2023 trough and the 2024-2025 AI-memory rebound.
- Memory peer cycle: 10 Micron year-to-year observations from FY2015-FY2025. Micron is not identical to SK hynix, but it faces similar DRAM/NAND pricing and capacity cycles, so it helps avoid treating one company history as the whole reference class.
- Analyst expectations with partial pooling: current published estimates are treated as noisy observations rather than ground truth. For FY2026 and FY2027, the model uses Hana Securities' quoted FnGuide consensus of about KRW 211.7tn and KRW 242.9tn, alongside Hana's own higher house view of KRW 229.4tn and KRW 286.7tn.
- General company-growth evidence: the simulation uses fat-tailed shocks and lower percentage volatility at larger scale, consistent with empirical findings that company growth rates are noisy, heavy-tailed, and less volatile for larger firms.
How overconfidence is reduced
Some observations overlap conceptually: SK hynix and Micron both sit inside the same memory cycle, and the near-term analyst path already reflects recent HBM demand. The model therefore does not count these as independent evidence streams. It gives the SK hynix and Micron cycle histories a shared memory-cycle layer, and then partially pools the structural path with analyst expectations using inverse-variance weights rather than snapping to a broker target. Near-term outside-view guidance error priors are set tighter than structural cycle error (about 0.075 log-sd in FY2026 and 0.105 in FY2027), so guidance gets more weight where its predictive power is stronger. In the current fit, the analyst layer gets an average weight of about 87% in FY2026 and 77% in FY2027. The fitted memory-cycle sample has an average annual log-growth equivalent of about 13%, a log-growth standard deviation of 0.37, average boom observations around 53%, and average bust observations around -32%.
Most influential historical examples
| Example | Revenue change | Why it matters | Context |
|---|---|---|---|
| SK hynix FY2025 | KRW 66.2tn to KRW 97.1tn, +47% | Latest anchor; HBM revenue more than doubled. | FY2025 results |
| SK hynix FY2024 | KRW 32.8tn to KRW 66.2tn, +102% | Shows how quickly memory revenue can rebound from a trough. | FY2024 results |
| SK hynix FY2023 | KRW 44.6tn to KRW 32.8tn, -27% | Downturn example that prevents straight-line AI extrapolation. | FY2024 comparative table |
| Micron FY2025 | $25.1bn to $37.4bn, +49% | Independent memory peer experiencing the same AI-led upcycle. | Micron FY2025 results |
| Micron FY2023 | $30.8bn to $15.5bn, -49% | Large peer-cycle bust used to keep downside tails wide. | Micron comparative figures |
Model structure
- Current company layer: starts from FY2025 revenue and 4Q25 run rate, then applies a decaying HBM uplift.
- Memory-cycle layer: draws from SK hynix and Micron historical boom and bust behavior, with persistent cycle shocks.
- Near-term analyst layer: infers a latent analyst target for FY2026-FY2027 from current consensus and house estimates, then partially pools that target with the structural path using precision weights.
- Simulation: 90,000 Monte Carlo paths produce yearly revenue samples. Each year's market ladder is
P(revenue >= threshold)at 50 friendly-rounded KRW trillion thresholds.
Forecast summary
| Fiscal year | Median | P10 | P90 |
|---|---|---|---|
| FY2026 | ₩212tn | ₩181tn | ₩249tn |
| FY2027 | ₩269tn | ₩208tn | ₩347tn |
| FY2028 | ₩308tn | ₩185tn | ₩489tn |
| FY2029 | ₩312tn | ₩154tn | ₩581tn |
Source notes
- SK hynix FY2025 results: FY2025 revenue of KRW 97.1467tn, FY2024 comparison of KRW 66.1930tn, 4Q25 revenue of KRW 32.8267tn, and HBM/HBM4 commentary.
- SK hynix FY2024 results: FY2024 and FY2023 revenue history, plus HBM share of DRAM revenue in 4Q24.
- SK hynix FY2022 results: direct context for the prior memory downturn.
- Micron FY2025 results: memory-peer cycle reference, including FY2025, FY2024, and FY2023 revenue figures.
- Hana Securities sector report: quoted FnGuide consensus and Hana house estimates for SK hynix FY2026-FY2027 revenue.
- Korea JoongAng Daily on 1Q26 consensus: contemporaneous FnGuide market consensus of KRW 46.6tn for SK hynix 1Q26 revenue.
- Stanley et al. (1996) and Bottazzi & Secchi (2006): motivation for fat-tailed growth shocks and size-sensitive volatility.
Data and reproducibility assets
Liquidity over time
i Shows the market’s available liquidity over time. Liquidity can decay continuously and may change discretely if liquidity adjustments are applied.Open this panel to load liquidity history.