AI Economic Scenarios 2030: Interactive Model

Explore the scenarios

How the model works

The economy produces output from many task instances combined with a constant-elasticity (CES) aggregator, with elasticity of substitution $σ$ (0.5 by default, so tasks are complements). A share $m_{t}$ of the economy's labor tasks is exposed to AI (cognitive work). Within that, a share $d_{t}$ of instances is actually done with AI, so $m_{t} d_{t}$ is the effective AI share. Each AI instance is $e^{a_{t}}$ times more productive. Of the AI-performed work, a share $ψ$ is automated (done by capital) and the rest is augmented (a worker using AI). Each automated instance creates $ρ$ new labor tasks (reinstatement). Capital is supplied with elasticity $ε$ = 3 with respect to its rental rate.

A note on notation. In the paper and on this page, $ψ$ is the automation share and $ρ$ is reinstatement. The share of work performed with AI is the product $m d$ . It is not a separate parameter.

Paths over time (paper)

m_{t} = \frac{\bar{m}}{1 + e^{- κ_{m} (t - t_{m})}}, κ_{m} = \frac{1}{3.5} \ln [\frac{\bar{m} - m_{2026}}{m_{2026}} \cdot \frac{m_{2030}}{\bar{m} - m_{2030}}]

The same logistic form applies to $d_{t}$ with ceiling 1. The mid-2026 anchors are $m$ = 0.14 and $d$ = 0.10, and the exposure ceiling is $\bar{m}$ = 0.624 (cognitive share of labor income). Productivity rises linearly: $a_{t} = a_{2026} + g_{a} (t - 2026.5)$ .

Closed-form solution (paper, Proposition 1, uniform parameters)

s_{L, t} = 1 - [s_{K 0} + s_{L 0} ψ m d (e^{- (1 - σ) a} - ρ)] e^{(1 - σ) x}

{\tilde{ℓ}}_{N} = - \ln (1 - m d [1 - ρ ψ - (1 - ψ) e^{- (1 - σ) a}])

Δ \ln w = \frac{Δ \ln s_{L} + {\tilde{ℓ}}_{N}}{1 - σ}, Δ \ln (Y / L) = Δ \ln w - Δ \ln s_{L}

ε x = Δ \ln K = \ln \frac{1 - s_{L}}{s_{K 0}} + Δ \ln (Y / L) - x

Here $x = Δ \ln r$ is the gap in the rental rate of capital. The page solves for it numerically by bisection (the root is unique). The base year is 2024, with labor share $s_{L 0}$ = 0.60. The cognitive-employment target is ${\tilde{ℓ}}_{C} \approx - (s_{N 0} / s_{C 0}) {\tilde{ℓ}}_{N}$ .

WealthLanding's illustrative additions

Unemployment. The paper models search and matching with sticky wages. This page uses a simpler stand-in. When the cognitive-employment target falls, the displaced workers join an excess pool $P_{t}$ . Each month a fraction $h / 12$ of the pool is rehired, so $U_{t} = \bar{U} + P_{t}$ with $\bar{U}$ = 3.8%. The preset speeds (1.1, 1.3 and 0.25 per year) were tuned so that 2030 unemployment lands near the paper's published values.
GDP with unemployment. $Δ \ln Y \approx Δ \ln (Y / L) + s_{L 0} \ln \frac{1 - U_{t}}{1 - \bar{U}}$ . Growth in year $t$ is the 12-month change in this gap plus the 2% baseline.
Power and chip caps. If power or chips can deliver only a fraction $c$ of what the scenario needs by 2030, the rise in $d$ and $a$ above their mid-2026 levels is scaled down. The scaling phases in linearly, reaching a factor of $c$ = min(power, chips) in 2030.
Compute tax. This scales the same rise by $1 - τ$ , using an assumed adoption elasticity of 1.
Investment feedback. Diffusion is multiplied by $1 + \min (0.25, φ \cdot {gap}_{t - 1})$ , capped at $d$ ≤ 1. A larger economy funds more AI deployment, and the bound prevents runaway paths.
Retraining. This raises the reabsorption speed $h$ by the chosen percentage.
Transfers/UBI. Worker household income = labor income + T × (capital income above the no-AI level). It is shown relative to baseline labor income.
Hyperscaler signal. Signal S is the average of two parts, each clipped to [−1, 1]: (average cloud growth − 20) / 40, and (CapEx growth year over year − 20) / 80. When the toggle is on, $d_{2030} \leftarrow d_{2030} (1 + 0.2 \cdot S \cdot strength)$ . If you enter an AI revenue share, it also nudges $a_{2030}$ by up to ±0.10. The reference points are WealthLanding judgment calls, not estimates.

Assumptions and limits

The paper's model, and this one, leave out robotics, aggregate-demand effects and an intertemporal saving (Euler) decision. The paper's ideas channel is small, and it is omitted here.
The paper allows parameters to differ across occupations. This page uses one uniform value for each parameter.
The average wage here is the full-employment wage from the closed form. The paper's figure also reflects wage rigidity, so ours differs, most visibly in the Extreme preset.
Results are gaps relative to a no-AI path with 2% growth. They are not forecasts.

Preset values (paper Table 1) and published 2030 results (Table 3)

Source: Korinek et al. (2026), Tables 1 and 3. Reabsorption speeds are WealthLanding's calibration.
Parameter / result	Modest	Substantial	Extreme
m₂₀₃₀	0.20	0.30	0.50
d₂₀₃₀	0.20	0.40	0.60
a mid-2026 → 2030	0.30 → 0.30	0.35 → 0.45	0.45 → 0.80
ψ (automation share)	0.50	0.75	0.90
ρ (reinstatement)	0.50	0.25	0
GDP above no-AI, 2030	1.6%	8.3%	32.4%
GDP growth, 2030	2.4%	5.4%	15.4%
Labor share, 2030	59.4%	56.1%	45.2%
Unemployment, 2030	3.9%	4.6%	11.9%

What it means for you

Career

In every preset, the pressure falls on cognitive tasks that AI can perform end to end. The model's key lever is reinstatement (ρ). Roles that keep creating new tasks for people hold up better. Building skills that sit next to AI, such as judgment, client relationships, and supervising AI output, matches the augmentation side of ψ. A larger emergency fund is a hedge against the slow-reabsorption case.

Investing themes

Higher-AI scenarios move income from labor to capital and raise the return on capital. They also lean on power, chips and data-center build-outs, which the caps above make explicit. These are themes for understanding the economy, not picks. Diversification still matters, because the same scenario that lifts AI-linked capital can hurt other sectors.

Policy

Try the policy sliders. Retraining shortens the unemployment hump. Transfers can keep worker household income at or above baseline even when the labor share falls. The paper estimates that a transfer of about 9% of GDP would keep cognitive workers' income at its no-AI level in the Extreme case. A compute tax slows adoption, which trades growth for a gentler transition.

Not investment advice. This page is an educational model. It is not a forecast, and it is not a recommendation to buy, sell or hold any security. Talk to a qualified professional about your own situation.

Sources

Korinek, A., Jones, C. I., Sacher, S., Cotter, K. and McCrory, A. (2026). Economic Scenarios for Transformative AI. The Anthropic Institute Working Paper No. 2026-02. PDF · CEPR DP21939 · Anthropic's scenario explorer
Hyperscaler inputs come from the quarter ended June 30, 2026: Amazon Q2 2026 (July 30, 2026) · Microsoft FY26 Q4 (July 29, 2026) · Alphabet Q2 2026 (July 22, 2026). Update them after each earnings release.

Embed this model

Add ?embed=1 to show only the interactive model, without the site header and footer:

<iframe src="https://www.wealthlanding.com/investment/ai-economic-scenarios.html?embed=1"
        title="AI Economic Scenarios 2030 (WealthLanding)" width="100%" height="1500"
        style="border:0" loading="lazy"></iframe>