The forecast, rebuilt from its own method
TR-2026-25 prices the energy-first transition to 2035 and states every assumption in the open. That is an invitation to rebuild it, so this page does: the model runs here, from the review's published inputs, and shows you where the reconstruction agrees with the published table and where it cannot.
What does relaxation-native compute save, and when?
Start from an AI data-centre load of 190 TWh in 2026 growing 22% a year to 2030 and 12% a year after that. Some share of that work is amenable to relaxation-native compute. On that share, efficiency improves by some multiple. Adoption follows an S-curve. Those three numbers are yours to set.
Does the reconstruction match the published table?
This is the part worth the page. The review publishes a headline table for three scenarios, so a rebuild can be checked against it rather than trusted. Pick a scenario preset above and this compares row by row.
| Row | Rebuilt here | Published | Agrees |
|---|
What the review would need to publish to make the rest reproducible
Twelve of the fifteen published cells rebuild exactly from the stated method: the 2030 and 2035 savings, the capital avoided and the carbon avoided, in all three scenarios. The three that do not are the cumulative totals over 2026 to 2035, and they cannot, for a specific and fixable reason.
Point-in-time results depend only on the adoption level in that year, and the review states those levels. A cumulative total depends on the adoption path through every intervening year, and the review states that the curve is an S-curve without stating its shape. Anchoring a logistic on the two published levels reproduces the endpoints exactly and the cumulative totals to within about 8%, and no choice of asymptote in that family fixes all three scenarios at once. One published parameter (the curve's midpoint year, or the adoption level in any single intervening year) would close it. That is a smaller thing to ask for than a workbook, and it would make the whole forecast reproducible by anyone.
Inputs and the published headline table are read from The Energy-First Turn (TR-2026-25), section on pricing the transition. The 84-entry evidence base behind it is at the energy-first database. Constraint vocabulary: the eight.