Anthropic Quantifies AI's Economic Impact: GDP Up 32% in Extreme Scenario but Knowledge Worker Wages Drop Over 10%
nashnova research
Anthropic's economics team released an interactive model mapping three scenarios for AI's impact on the U.S. economy — even under the mildest assumptions, capital's share of GDP rises and knowledge-worker wages face downward pressure, meaning the question of who captures AI's gains has moved from speculation to a modelable structural risk.
What is this model actually calculating?
Anthropic, working with economists Anton Korinek and Chad Jones, broke every job into discrete tasks and scored each one: can AI automate it, augment it, leave it untouched, or create entirely new work?
In plain terms = instead of asking "will AI replace you?" in the abstract, the model slices your job into dozens of pieces and grades them one by one.
It then runs three scenarios — mild, substantial, extreme — corresponding to low-to-high AI penetration rates.
How wide is the gap between the three scenarios?
Mild scenario: impact resembles the internet era. GDP rises just 1.6% above baseline; unemployment stays in the normal range.
Substantial scenario: by 2030 AI handles roughly half of knowledge work autonomously. GDP is 8.3% above baseline at about $36 trillion. Overall wages rise 2.1%, but knowledge-worker wages fall 0.3% while other workers' wages rise 5.9%.
Extreme scenario: AI outperforms humans on nearly every knowledge task. Annual GDP growth hits 15%, the economy roughly doubles every four and a half years, and GDP reaches about $44.4 trillion — 32% above baseline. But knowledge-worker wages drop over 10%, with unemployment far exceeding a typical recession.
This means → the GDP spread across the three paths is 1.6% vs. 32%, yet the direction is the same: the higher AI climbs, the less knowledge workers win.
Why does capital's slice of the pie keep growing?
Labor's share of GDP currently sits at roughly 60%. In the substantial scenario it falls to 56.1%; in the extreme scenario it drops to 55% — capital's share rises to 43.9% and 45% respectively.
This reflects a structural signal that runs across all three scenarios: capital's rising share is not unique to the extreme case — it shows up in the mild one too.
In plain terms = fast or slow, the share of total output flowing to "people who own the machines" grows, while the share flowing to "people who work with their brains" shrinks.
Who actually gets a raise?
Even in the substantial scenario where overall wages rise, the gains concentrate in construction and other non-knowledge occupations, not among knowledge workers.
This means → AI's aggregate wage boost is an averages trap — the total goes up, but if you code, analyze, or design for a living, your slice may be shrinking.
Anthropic surveyed over 10,000 Americans alongside the model. The median respondent's expectation aligns closest with the substantial scenario; roughly one in ten holds views consistent with the extreme case.
What is the model still missing?
Anthropic labels it a version 1.0, explicitly noting it does not yet account for business-cycle fluctuations, policy responses, or humanoid robots.
This means → all three scenarios model a "pure AI shock" — no government reaction, no prior recession baked in — so the actual path will almost certainly be more complex.
Anthropic says the model will guide the policy interventions it funds going forward. Whether capital's share can keep expanding, and how fast and forcefully policy responds, are the key variables that will determine which scenario path holds.
市场有风险，内容仅供研究参考，不构成投资建议。