Forecast a Binary Question End to End
Binary mode returns a calibrated probability from 0 to 100 that a yes/no question resolves YES, plus a rationale with citations. One question, one row, one call.
The question
| Field | Value |
|---|---|
| Question | Will AWS, Azure, Google Cloud, or Oracle announce a deployment of Intel's Crescent Island GPU by December 31, 2027? |
| Outcome type | binary |
| Resolves | A public announcement from any of the four clouds naming a Crescent Island deployment, on or before December 31, 2027 |
Crescent Island is Intel's inference GPU built on LPDDR5X rather than HBM.
Run it
import asyncio
from pandas import DataFrame
from futuresearch.ops import forecast
async def main():
result = await forecast(
input=DataFrame([
{
"question": (
"Will AWS, Azure, Google Cloud, or Oracle announce a deployment "
"of Intel's Crescent Island GPU by December 31, 2027?"
),
"resolution_criteria": (
"Resolves YES on a public announcement from any of AWS, Azure, "
"Google Cloud, or Oracle naming a Crescent Island deployment, on "
"or before December 31, 2027."
),
"background": (
"Crescent Island is Intel's LPDDR5X-based inference GPU. Intel's "
"two prior Gaudi generations won no top-four cloud deployments."
),
},
]),
forecast_type="binary",
effort_level="HIGH",
)
print(result.data[["question", "probability", "rationale"]])
asyncio.run(main())
probability is clamped to the range 3 to 97, so near-certain outcomes keep some residual uncertainty.
Results
| Question | Probability |
|---|---|
| A top-four cloud announces a Crescent Island deployment by December 31, 2027 | 13% |
The anchor in the rationale is the base rate: zero top-four wins across two Gaudi generations.
What would move the number up:
- Oracle building a budget inference tier on a non-Nvidia accelerator, which it has done for price-sensitive workloads before
- A deep co-engineering partnership announced at re:Invent or the OCP Summit, bypassing the slow validation pipeline
- Independent testing showing unprecedented cost-per-token efficiency
- Intel disclosing strong, independently verified inference economics in late 2026
- Any named cloud co-engineering commitment
Going deeper
- Long-form analysis: Intel Crescent Island: Why No Major Cloud Will Deploy It by 2027
- Published output: the full forecast and its sources
- Guide: Turn Claude into an accurate forecaster