Dashed green is what AI 2027 wrote down for each date; black is what has been measured. The vertical line is the last day the real data was reviewed.
Length of software task (in human time) a frontier model completes half the time. The scenario assumes doubling every four months from the spring-2025 frontier and an 80% horizon of "years" by March 2027.
Best public score on the verified subset. The scenario set 85% by mid-2025 as its Agent-1 target.
How much faster AI makes AI research. The scenario's multiplier for OpenBrain against the only measurement that exists: METR's randomized trial of experienced developers (19% slower in 2025; a 2026 follow-up found no reliable speed-up).
Annualized run-rate revenue. OpenBrain in the scenario; OpenAI and Anthropic as reported.
Post-money valuation at each funding round or tender. OpenBrain in the scenario; OpenAI and Anthropic as reported.
Spending per calendar year, plotted at year end: the scenario's global datacenter figure against the five largest hyperscalers' capital expenditure.