Macro / GDP & AI Scenarios ยท Saved data snapshot: October 5, 2026
Assumptions behind this chart
- Projections are frozen as of 2026-10-05. All three projected paths are saved in projections_frozen_2026-10-05.csv and are not recalculated on later runs. Later BEA releases, including revisions to past quarters, move only the blue reported line, so the chart shows how actual GDP comes in against fixed October 2026 expectations. Once BEA reports a projected quarter, its frozen growth rates appear as dots on the blue bar. Deleting the frozen files re-freezes the paths from the data available on that day.
- Common starting point. Every path starts from Q2-26 real GDP as reported on 2026-10-05: $24.41tn in chained 2017 dollars at a seasonally adjusted annual rate, after 2.2% growth that quarter.
- Baseline, gray line. Q3-26 uses the Atlanta Fed GDPNow estimate of 3.7% as of 2026-10-01. Every later quarter grows 2.7% at an annual rate, the average real GDP growth from 2023 through Q2 2026 (2012 to 2019 averaged 2.5%). It reaches $27.58tn in Q4-30.
- Paper's no-AI path, light gray line. This is the economy without AI in Korinek, Jones, Sacher, Cotter and McCrory (2026), Economic Scenarios for Transformative AI, Anthropic Institute Working Paper 2026-02. It grows 2.0% a year: 1.67% from new ideas and 0.33% from labor-force growth. It starts 0.7% below Q2-26 reported GDP, because the paper's AI gap had already opened by then, and it reaches $26.52tn in Q4-30.
- Substantial change scenario, orange line and bars. This is the paper's no-AI path multiplied by the paper's AI gap. The gap comes from a reconstruction of the paper's model at its substantial-scenario settings: by 2030 AI can affect 30% of tasks, is used on 40% of them, and makes each AI-done task 0.45 log points more productive; 75% of AI use automates work, and one new human task appears for every four automated. The gap is scaled so GDP is exactly 8.3% above the no-AI path on Jan 1 2030, which gives 5.4% growth over 2029, both as in the paper's Table 3. The paper stops at Jan 2030, so Q2 to Q4 2030 are held at 5.4%. Growth starts at 2.6% in Q3-26 and rises over time, because AI capability and adoption follow S-shaped curves. It reaches $29.43tn in Q4-30.
- GDPNow, green. Atlanta Fed GDPNow nowcasts come from FRED and its archive of past releases. Small green dots show each quarter's final nowcast before BEA's release, and the large dot is the latest nowcast for the quarter now being tracked. The bottom panel shows every daily update since January 2025.
- Index chart, below the main charts. Each line is divided by its own Q2-26 level and multiplied by 100. The reported line uses whatever BEA publishes on the day the chart is run, divided by that same day's figure for the starting quarter. BEA moves the base year of its real GDP dollars about every five years; the last move, to 2017 dollars, came in 2023, and the next is likely around 2028. That shifts every dollar level at once but cancels out of the index. The September annual revisions cancel out the same way. The frozen paths are divided by the frozen starting level, so they never move.
- Gap bar charts, bottom: reported minus substantial, then reported minus baseline. Each bar is reported real GDP minus that path for the quarter, in billions of chained 2017 dollars. Blue means reported GDP is above the path and red means below. Reported GDP is first rescaled to the October 2026 data, as current level times the frozen starting level over today's starting level, so revisions and base-year changes don't create false jumps. Both gaps are zero in Q2-26 by construction. Before that quarter the paths are backcasts frozen in backcast_frozen_2026-10-05.csv. The substantial backcast runs the same scenario formula backward. The baseline backcast runs its 2.7% rate backward. Bars for 3Q 2026 to 4Q 2030 appear as BEA reports each quarter. The faded green bar is the gap GDPNow implies for the next quarter due.
- Caveats. The scenarios are illustrative, not forecasts. The paper models what the economy can produce and leaves out business cycles and demand swings, so reported quarters will scatter around any of these paths. Growth rates are quarter-on-quarter at a seasonally adjusted annual rate. The paper quotes its 5.4% as a log change, which equals 5.5% as an ordinary percentage.