Forecast Track Record
Every month, on the 2nd, our live scenario models' bear / base / bull projections are frozen into write-once vintage rows. A projection changed before it was frozen is gone — the ledger cannot be backfilled. When a target period matures, the row is graded against admissible published benchmark observations (see methodology) and the result stays on the record: correct, partial or incorrect, with the signed error.
No row has been graded yet. 18 rows have matured without a grade: 9 are in a series with no mapped benchmark; 9 are still waiting for two admissible readings from independent sources. They stay on the page ungraded rather than scored against a mismatched reading — a grade is permanent, so declining is the default.
HBM Price Scenario Model
hbm-pricing-model · v1.1Projects HBM3E and HBM4 contract pricing per GB through Q4 2027 (Q2 and Q4 2027 targets), on a per-stack contract basis at a 12-Hi 36GB reference: settled 2026 per-stack prices, the 2027 renewal repricing, and a second-half step.
Revised 2026-10-04 (v1.1): Restated on a per-stack contract basis ($/GB at a 12-Hi 36GB stack) with 2027 targets only, replacing v1.0's oversupply-driven 2026 declines after 2026 renewals rose about 20% and 2027 repricing was reported. Each row below keeps the version it was frozen under.
| Series | Scenario | Target period | Predicted (band) | Unit | Vintage | Status |
|---|---|---|---|---|---|---|
| HBM3E contract price | base | Q4 2026 | 9 (7.5–10.5) | USD/GB | 2026-10 · v1.0 | matures 2026-12-31 |
| HBM3E contract price | bear | Q4 2026 | 6 (4.5–8) | USD/GB | 2026-10 · v1.0 | matures 2026-12-31 |
| HBM4 contract price | base | Q4 2026 | 13 (11–15) | USD/GB | 2026-10 · v1.0 | matures 2026-12-31 |
| HBM4 contract price | bear | Q4 2026 | 11 (9–13) | USD/GB | 2026-10 · v1.0 | matures 2026-12-31 |
| HBM4 contract price | bull | Q4 2026 | 15.5 (13–18) | USD/GB | 2026-10 · v1.0 | matures 2026-12-31 |
| HBM3E contract price | base | Q3 2026 | 9.5 (8–11) | USD/GB | 2026-09 · v1.0 | matured 2026-09-30 · awaiting admissible data |
| HBM3E contract price | bear | Q3 2026 | 7 (5.5–9) | USD/GB | 2026-09 · v1.0 | matured 2026-09-30 · awaiting admissible data |
| HBM3E contract price | bull | Q3 2026 | 17 (14–20) | USD/GB | 2026-09 · v1.0 | matured 2026-09-30 · awaiting admissible data |
| HBM3E contract price | base | Q4 2026 | 9 (7.5–10.5) | USD/GB | 2026-09 · v1.0 | matures 2026-12-31 |
| HBM3E contract price | bear | Q4 2026 | 6 (4.5–8) | USD/GB | 2026-09 · v1.0 | matures 2026-12-31 |
| HBM4 contract price | base | Q4 2026 | 13 (11–15) | USD/GB | 2026-09 · v1.0 | matures 2026-12-31 |
| HBM4 contract price | bear | Q4 2026 | 11 (9–13) | USD/GB | 2026-09 · v1.0 | matures 2026-12-31 |
| HBM4 contract price | bull | Q4 2026 | 15.5 (13–18) | USD/GB | 2026-09 · v1.0 | matures 2026-12-31 |
| HBM3E contract price | base | Q3 2026 | 9.5 (8–11) | USD/GB | 2026-08 · v1.0 | matured 2026-09-30 · awaiting admissible data |
| HBM3E contract price | bear | Q3 2026 | 7 (5.5–9) | USD/GB | 2026-08 · v1.0 | matured 2026-09-30 · awaiting admissible data |
| HBM3E contract price | bull | Q3 2026 | 17 (14–20) | USD/GB | 2026-08 · v1.0 | matured 2026-09-30 · awaiting admissible data |
| HBM3E contract price | base | Q4 2026 | 9 (7.5–10.5) | USD/GB | 2026-08 · v1.0 | matures 2026-12-31 |
| HBM3E contract price | bear | Q4 2026 | 6 (4.5–8) | USD/GB | 2026-08 · v1.0 | matures 2026-12-31 |
| HBM4 contract price | base | Q4 2026 | 13 (11–15) | USD/GB | 2026-08 · v1.0 | matures 2026-12-31 |
| HBM4 contract price | bear | Q4 2026 | 11 (9–13) | USD/GB | 2026-08 · v1.0 | matures 2026-12-31 |
| HBM4 contract price | bull | Q4 2026 | 15.5 (13–18) | USD/GB | 2026-08 · v1.0 | matures 2026-12-31 |
| HBM3E contract price | base | Q3 2026 | 9.5 (8–11) | USD/GB | 2026-07 · v1.0 | matured 2026-09-30 · awaiting admissible data |
| HBM3E contract price | bear | Q3 2026 | 7 (5.5–9) | USD/GB | 2026-07 · v1.0 | matured 2026-09-30 · awaiting admissible data |
| HBM3E contract price | bull | Q3 2026 | 17 (14–20) | USD/GB | 2026-07 · v1.0 | matured 2026-09-30 · awaiting admissible data |
| HBM3E contract price | base | Q4 2026 | 9 (7.5–10.5) | USD/GB | 2026-07 · v1.0 | matures 2026-12-31 |
| HBM3E contract price | bear | Q4 2026 | 6 (4.5–8) | USD/GB | 2026-07 · v1.0 | matures 2026-12-31 |
| HBM4 contract price | base | Q4 2026 | 13 (11–15) | USD/GB | 2026-07 · v1.0 | matures 2026-12-31 |
| HBM4 contract price | bear | Q4 2026 | 11 (9–13) | USD/GB | 2026-07 · v1.0 | matures 2026-12-31 |
| HBM4 contract price | bull | Q4 2026 | 15.5 (13–18) | USD/GB | 2026-07 · v1.0 | matures 2026-12-31 |
DDR4 Price Scenario Model
dram-ddr4-model · v1.0Projects DDR4 8Gb spot and contract pricing 2026–2027, driven by the EOL production decline curve, residual industrial/automotive demand, and DDR5 substitution rates.
| Series | Scenario | Target period | Predicted (band) | Unit | Vintage | Status |
|---|---|---|---|---|---|---|
| DDR4 8Gb spot price | base | Q4 2026 | 6 (4.5–8) | USD (DDR4 8Gb) | 2026-10 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| DDR4 8Gb spot price | bear | Q4 2026 | 3 (2–4.5) | USD (DDR4 8Gb) | 2026-10 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| DDR4 8Gb spot price | bull | Q4 2026 | 10 (7–13) | USD (DDR4 8Gb) | 2026-10 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| DDR4 8Gb spot price | base | Q3 2026 | 6.5 (5–8.5) | USD (DDR4 8Gb) | 2026-09 · v1.0 | matured 2026-09-30 · no mapped benchmark |
| DDR4 8Gb spot price | bear | Q3 2026 | 3.5 (2.5–5) | USD (DDR4 8Gb) | 2026-09 · v1.0 | matured 2026-09-30 · no mapped benchmark |
| DDR4 8Gb spot price | bull | Q3 2026 | 11 (8–14) | USD (DDR4 8Gb) | 2026-09 · v1.0 | matured 2026-09-30 · no mapped benchmark |
| DDR4 8Gb spot price | base | Q4 2026 | 6 (4.5–8) | USD (DDR4 8Gb) | 2026-09 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| DDR4 8Gb spot price | bear | Q4 2026 | 3 (2–4.5) | USD (DDR4 8Gb) | 2026-09 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| DDR4 8Gb spot price | bull | Q4 2026 | 10 (7–13) | USD (DDR4 8Gb) | 2026-09 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| DDR4 8Gb spot price | base | Q3 2026 | 6.5 (5–8.5) | USD (DDR4 8Gb) | 2026-08 · v1.0 | matured 2026-09-30 · no mapped benchmark |
| DDR4 8Gb spot price | bear | Q3 2026 | 3.5 (2.5–5) | USD (DDR4 8Gb) | 2026-08 · v1.0 | matured 2026-09-30 · no mapped benchmark |
| DDR4 8Gb spot price | bull | Q3 2026 | 11 (8–14) | USD (DDR4 8Gb) | 2026-08 · v1.0 | matured 2026-09-30 · no mapped benchmark |
| DDR4 8Gb spot price | base | Q4 2026 | 6 (4.5–8) | USD (DDR4 8Gb) | 2026-08 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| DDR4 8Gb spot price | bear | Q4 2026 | 3 (2–4.5) | USD (DDR4 8Gb) | 2026-08 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| DDR4 8Gb spot price | bull | Q4 2026 | 10 (7–13) | USD (DDR4 8Gb) | 2026-08 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| DDR4 8Gb spot price | base | Q3 2026 | 6.5 (5–8.5) | USD (DDR4 8Gb) | 2026-07 · v1.0 | matured 2026-09-30 · no mapped benchmark |
| DDR4 8Gb spot price | bear | Q3 2026 | 3.5 (2.5–5) | USD (DDR4 8Gb) | 2026-07 · v1.0 | matured 2026-09-30 · no mapped benchmark |
| DDR4 8Gb spot price | bull | Q3 2026 | 11 (8–14) | USD (DDR4 8Gb) | 2026-07 · v1.0 | matured 2026-09-30 · no mapped benchmark |
| DDR4 8Gb spot price | base | Q4 2026 | 6 (4.5–8) | USD (DDR4 8Gb) | 2026-07 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| DDR4 8Gb spot price | bear | Q4 2026 | 3 (2–4.5) | USD (DDR4 8Gb) | 2026-07 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| DDR4 8Gb spot price | bull | Q4 2026 | 10 (7–13) | USD (DDR4 8Gb) | 2026-07 · v1.0 | matures 2026-12-31 · no mapped benchmark |
CoWoS Capacity Expansion Model
cowos-capacity-model · v1.0Projects TSMC CoWoS monthly capacity through end 2027, driven by TSMC capex commitment, packaging tool lead times, and AI chip demand from hyperscalers.
| Series | Scenario | Target period | Predicted (band) | Unit | Vintage | Status |
|---|---|---|---|---|---|---|
| CoWoS total capacity | base | End 2026 | 106,000 (95,000–120,000) | wafers/month | 2026-10 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| CoWoS total capacity | bear | End 2026 | 88,000 (80,000–98,000) | wafers/month | 2026-10 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| CoWoS total capacity | bull | End 2026 | 128,000 (115,000–142,000) | wafers/month | 2026-10 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| CoWoS total capacity | base | End 2027 | 155,000 (135,000–175,000) | wafers/month | 2026-10 · v1.0 | matures 2027-12-31 · no mapped benchmark |
| CoWoS total capacity | bear | End 2027 | 110,000 (95,000–125,000) | wafers/month | 2026-10 · v1.0 | matures 2027-12-31 · no mapped benchmark |
| CoWoS total capacity | bull | End 2027 | 200,000 (175,000–225,000) | wafers/month | 2026-10 · v1.0 | matures 2027-12-31 · no mapped benchmark |
| CoWoS total capacity | base | End 2026 | 106,000 (95,000–120,000) | wafers/month | 2026-09 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| CoWoS total capacity | bear | End 2026 | 88,000 (80,000–98,000) | wafers/month | 2026-09 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| CoWoS total capacity | bull | End 2026 | 128,000 (115,000–142,000) | wafers/month | 2026-09 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| CoWoS total capacity | base | End 2027 | 155,000 (135,000–175,000) | wafers/month | 2026-09 · v1.0 | matures 2027-12-31 · no mapped benchmark |
| CoWoS total capacity | bear | End 2027 | 110,000 (95,000–125,000) | wafers/month | 2026-09 · v1.0 | matures 2027-12-31 · no mapped benchmark |
| CoWoS total capacity | bull | End 2027 | 200,000 (175,000–225,000) | wafers/month | 2026-09 · v1.0 | matures 2027-12-31 · no mapped benchmark |
| CoWoS total capacity | base | End 2026 | 106,000 (95,000–120,000) | wafers/month | 2026-08 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| CoWoS total capacity | bear | End 2026 | 88,000 (80,000–98,000) | wafers/month | 2026-08 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| CoWoS total capacity | bull | End 2026 | 128,000 (115,000–142,000) | wafers/month | 2026-08 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| CoWoS total capacity | base | End 2027 | 155,000 (135,000–175,000) | wafers/month | 2026-08 · v1.0 | matures 2027-12-31 · no mapped benchmark |
| CoWoS total capacity | bear | End 2027 | 110,000 (95,000–125,000) | wafers/month | 2026-08 · v1.0 | matures 2027-12-31 · no mapped benchmark |
| CoWoS total capacity | bull | End 2027 | 200,000 (175,000–225,000) | wafers/month | 2026-08 · v1.0 | matures 2027-12-31 · no mapped benchmark |
| CoWoS total capacity | base | End 2026 | 106,000 (95,000–120,000) | wafers/month | 2026-07 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| CoWoS total capacity | bear | End 2026 | 88,000 (80,000–98,000) | wafers/month | 2026-07 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| CoWoS total capacity | bull | End 2026 | 128,000 (115,000–142,000) | wafers/month | 2026-07 · v1.0 | matures 2026-12-31 · no mapped benchmark |
| CoWoS total capacity | base | End 2027 | 155,000 (135,000–175,000) | wafers/month | 2026-07 · v1.0 | matures 2027-12-31 · no mapped benchmark |
| CoWoS total capacity | bear | End 2027 | 110,000 (95,000–125,000) | wafers/month | 2026-07 · v1.0 | matures 2027-12-31 · no mapped benchmark |
| CoWoS total capacity | bull | End 2027 | 200,000 (175,000–225,000) | wafers/month | 2026-07 · v1.0 | matures 2027-12-31 · no mapped benchmark |
Methodology
- Freeze scope: quarterly and year-end target periods only, future periods only, cost and supply metrics only — never securities price targets. N2/2nm series are excluded at the freeze layer.
- Write-once: a database trigger rejects any edit to a frozen prediction; grading fields can be set exactly once. Corrections to our live models never rewrite a frozen vintage.
- Grading: a matured row is scored only against admissible published benchmark observations — the same memory generation as the series (an HBM3E forecast is never scored against HBM3 or HBM4 readings), the same price basis its model version declares (a per-stack contract price is never scored against a manufacturing cost or a blended average), a delivery period overlapping the target quarter (not merely published during it), and an external source: at least two independent hosts, one reading each, never our own pages or constants. The realized value is their median. Inside the frozen low–high band = correct; within 30% of the mid = partial; beyond = incorrect. With fewer than two admissible readings the row stays ungraded and is retried at the next monthly run.
- Price basis by model version: the HBM pricing model v1.0 never stated which price its $/GB forecasts are. On 2026-10-04, before any v1.0 row was graded and while no reading yet carried a basis tag, we ruled its HBM3E forecasts a revenue-implied average and its HBM4 forecasts a per-stack contract price. v1.1 states a per-stack contract price for both. A ruled row is gradeable, not graded: it still needs two admissible readings on its own basis.
- Not gradeable yet: a model version is graded only once it declares which price its forecasts are. Series without a mapped benchmark (CoWoS total capacity, DDR4 8Gb today) stay ungraded rather than being scored against a proxy.
- Machine-readable: /api/v1/track-record (JSON, public, full fidelity) and the
get_track_recordMCP tool.
Looking for what other organizations predicted? Third-party forecast vintages (TrendForce, WSTS, company guidance) are archived — not graded — at /data/forecast-vintages.