Invest1 publisher3 min readPublished
Anthropic brackets AI's economic impact between internet-sized gains and 15 per cent annual growth
Anthropic published three scenarios and an interactive model, declined to say which is likelier, and left aggregate demand out of the arithmetic. The same newsletter carrying the research reports business AI spending is falling.
The Investor · Invest desk
What happened
- Anthropic published three scenarios for AI's effect on the U.S. economy alongside an interactive model that lets anyone enter their own assumptions on capability, adoption speed and worker substitution.
- In the fastest scenario, AI does almost all knowledge work autonomously and creates essentially no new jobs, GDP grows 15 per cent a year, and unemployment climbs well past anything seen in a typical recession.
- In the slowest, AI is a helpful sidekick whose impact is roughly on par with the internet, producing real gains that arrive gradually, the kind of growth already seen before.
- Anthropic is not saying which of the three is more likely, and the company acknowledges it does not know exactly how AI will affect the economy.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint With the demand side switched off, the 15 per cent figure is a ceiling computed without the mechanism that would damp it, namely the consumer spending lost when incomes go.
- decision Anyone allocating capital against AI has to underwrite adoption timing, because the company's own public benefit lead puts diffusion slower than capability improvement.
- contradiction The scenarios reach to the end of the decade while the nearest data point in the same newsletter has business AI spend going down, so the near-term series and the long-run model point opposite ways.
- precedent Shipping an unweighted explorer sets the expectation that the vendor supplies the framework and outside economists supply the probabilities.
The third scenario puts GDP growth at 15 per cent a year while unemployment climbs well past anything seen in a typical recession, and the model behind it never asks who buys the output [4]. It calculates GDP purely from the supply side, from how much AI lifts productivity and output, without accounting for whether people who lost their income can still spend [5]. Anthropic's own tool says the scenario explorer "leaves out policy responses, business cycles, potential aggregate demand or financial market disruptions, and possible catastrophic risks", and calls the whole framework a "stark simplification of a complex reality" [6].
The middle path has AI doing half of all knowledge work by 2030, mostly autonomously, the economy growing at twice its normal rate, and knowledge workers' wages stalling while everyone else gains [3].
The compounding is loose too. Fifteen per cent a year for four and a half years multiplies output by 1.876, so the economy is 87.6 per cent bigger, not twice the size [15]; doubling at that rate takes ln 2 divided by ln 1.15, about 4.96 years [16]. Anthropic said its aim is to give economists and policy experts concrete scenarios about where we might be by the end of the decade [8].
Jack Clark, the cofounder who is head of public benefit and leads the Anthropic Institute [17], told NPR he expects the technology itself to keep improving "at a very, very fast and sustained rate" but thinks it will spread through the economy "more slowly" than most people assume [9]. He also told NPR that the tax windfall from that growth could give policymakers room to help displaced workers, something "unimaginable today" [10].
The nearest-term reading in the same newsletter runs the other way: Fortune lists falling business AI spend among the day's items, with no figure attached to it [11]. Economists Ben Moll and Alex Imas dismissed predictions of double-digit growth in a new essay, according to Fortune, whose account of it breaks off mid-sentence [14]. Anthropic's companion survey of nearly 11,000 people found the public expecting real productivity gains alongside pain for workers in AI-exposed jobs, with rising worry about younger generations and entry-level roles [12].
My read: a range running from internet-sized to 15 per cent a year is a company telling you it cannot size the market for its own product, and the interactive tool hands that job to whoever wants it [1][2][4]. The counter-thesis has force. When the swing variable is other people's adoption decisions, an unweighted range is the honest output, and more than 200 economists, executives and researchers, Stiglitz and Krugman and Acemoglu among them, asked in July for this kind of work before a transformation they called "larger than the Industrial Revolution" [13]. The release does not close the demand loop [5], and it does not weight the paths [7]. I would give up this reading the day Anthropic attaches a probability to any of the three [7].
What to watch
- Whether Fortune or Anthropic puts a figure behind the falling business AI spend line, and over what period it is measured.
- Whether the interactive explorer gains a demand-side block, which would change the third scenario's 15 per cent output.
- Whether Anthropic answers the Moll and Imas essay dismissing double-digit growth predictions.