15 September 2026
Heard In AI

Anthropic's fastest growth scenario grows the pie and shrinks labor's slice

On Moonshots, the panel read out Anthropic's most extreme economic scenario: roughly 15% annual GDP growth by 2030 alongside 17.9% unemployment among cognitive workers and labor's share of income falling from about 60% to 45%. Two investors called the growth number a lowball. Emad Mostaque said the model is missing the thing that breaks it — aggregate demand — and the panel fell into an argument about dividends, ownership and who pays the displaced.

A briefing reports one development when it happens. We correct or clarify it later; a new development gets a new briefing. How our formats work

The host had a set of numbers and a question about arithmetic. Anthropic's economic impact report, he told the Moonshots panel, puts forward three futures. In the most extreme one, AI performs almost half of today's cognitive work by 2030, GDP growth reaches 15% a year, the labor share of income falls from 60% to 45%, and nearly one in five cognitive workers is unemployed.

"15% growth means the pie gets enormous," he said. "A 45% labor share means the way we slice it stops working."

That is the fork the rest of the segment argued over: not only whether the growth is plausible, but whether an economy can grow that fast while the people who would have bought its output are losing their jobs.

What the scenarios actually are

Anthropic's scenario explorer, published in September 2026, is not a forecast. It is an interactive model that lets you set assumptions about how capable AI becomes and how fast it is adopted, then shows what follows through 2030. It treats jobs as bundles of tasks, drawn from the US government's O*NET occupational database. For a nurse, AI might draft discharge instructions, automate supply ordering and create new work reviewing AI-proposed care plans, while the physical side of care stays where it is.

The technical working paper behind it, by Anton Korinek, Charles I. Jones, Szymon Sacher, Tess Cotter and Peter McCrory, simulates 2026 to 2030 month by month, splitting the workforce into cognitive and other occupations and tracking automation, augmentation, innovation, how quickly capital can be built and how long it takes people to find new jobs.

The milder of the two headline cases — the substantial-change scenario the host called "moderate" — puts 2030 GDP 8.3% above a world with no further AI progress, with cognitive employment down 3.9% from mid-2026. The extreme scenario is the one with the 15% annual growth rate, a GDP level 32.4% above the no-AI baseline, 17.9% unemployment among cognitive workers against 11.9% for all workers, and the fall in labor's share.

One line in the paper is the one the panel fought about. Total labor income stays roughly unchanged relative to the no-AI path, because higher incomes outside cognitive occupations offset the losses inside them. Most of the aggregate gains go to capital — to whoever owns the machines and the companies. The paper's own sensitivity tests show the picture is not fixed: constraints on how fast capital can be supplied, and wages that do not adjust downward easily, materially change wages and unemployment. Most of the divergence between scenarios happens after 2027.

Version 1.0 also says what it leaves out: hyper-capable robots, policy responses, business cycles, aggregate-demand or financial-market disruptions, and catastrophic risks.

The panel thinks the number is too small

Dave's objection to 15% was that it is not enough. The legacy investment world, he said, points at a hundred years of history and says nothing like this ever happens — except that industrializing China did every bit of it, to which the answer is always that China started from a low base and was catching up. "But look, at the end of the day, we've never experienced anything vaguely like the singularity before. And the playbooks have to get thrown out." As an investor of more than thirty years, he said, he believes the scenario "is, if anything, a lower bound."

The host added a live data point: the Atlanta Fed's GDPNow tracker, which estimates growth in real time, had third-quarter US growth at about 4.7% annualized against 1.5% in the second quarter, driven by consumer spending and private investment. More than double the long-run average, he said — the kind of number seen coming out of the pandemic, except that this time there is no reopening, only capital spending.

Alex went further still and questioned the ruler rather than the reading. He thinks Anthropic is lowballing, and that even the regional Feds lack the instruments to measure what is coming; he expects real wealth to double or triple year over year deeper into the singularity. The official measures, in his telling, may do something stranger than rise: "Maybe it looks like 15%. Maybe it looks negative," like "a compass needle going around in circles when you're near a magnetic pole."

Mostaque: the model is missing the demand side

Emad Mostaque, who has written a book on the subject, took the opposite view of the same numbers. "We're screwed with these numbers," he said. "And they have not done their model properly." His reason was specific: "There's a complete collapse in aggregate demand."

Aggregate demand is simply the total spending in an economy — households, businesses and government buying things. Mostaque's objection is that the paper's own result depends on it holding up. He accepts the supply side of the model: the returns flowing to capital, he said, look correct. What he rejects is aggregate wages staying constant, which he called mathematically impossible alongside the rest. He was careful to praise the authors — "this is from Anthropic and great guys like Anton Koronek and others" — while calling it "a good model for half of it." The paper's stated exclusions include exactly the thing he names.

His own model, he said, will be released in a few weeks at ii.inc, open-sourced, built around the idea that intelligence is becoming reproducible capital rather than something scarce inside human workers. The urgency he attached to it was about people rather than parameters: 20% of cognitive workers unemployed in three or four years, he said, and truck drivers when the robots arrive on a similar timetable. "We need to catch them." Even on the moderate scenario, he argued, the economic disruption is as big as COVID, and the human disruption bigger. Governments, in his view, have to prepare now to support the people who fall through the nets.

"Not anywhere close to rocket science"

Alex offered the counterpoint. The capital means of production are still held by humans, he said. The scenario he would call doom-adjacent is the one where that changes — where "biological meat, body humans get economically disenfranchised by AIs" and the AIs trade with each other instead. Short of that, he said, the problem is old and the tools already exist: "Dividends, sovereign wealth funds, UBE, UBI. I don't think any of this is anywhere close to rocket science if we find ourselves in a scenario of extreme capital accumulation due to superintelligence."

Mostaque agreed the gains go to the GPU owners and that mechanisms exist — he has proposed his own, alongside UBI — but kept returning to the disruption itself rather than the remedy's difficulty. The host's worry was political: "It doesn't take a lot of angry, angry young men who haven't got a job, you know, can't afford a house and a car, can't get married to start a revolution."

Alex said the message has been received, particularly in Western countries where the capital is concentrating, pointing to what he described as the president previewing a proposed universal basic dividend of $5,000 per person in the previous 24 hours — "presumably a singularity dividend." He does not regard any of this as existential. The host preferred his own earlier prediction of $3,000 a month, against roughly $1,000 a month during COVID, and argued that as AI and robotics drive costs down, that sum eventually buys what he calls universal high income.

The segment ended on a different way of assigning the bill. The host took the fight over data centers as a case study: neighbors objected that a data center would raise their power prices, and the resolution was to require anyone building one to drive the local cost of power down instead — how was up to them. Apply the same rule, he said, to a company sitting inside a $30 trillion addressable market with 20% of knowledge work at risk. For a firm with that much capital and opportunity, not eliminating those jobs is "like a rounding error of effort. You just make it an obligation to their own success and the problem goes away."

Share this article

Go to the original

Sources & further reading

  1. 01
  2. 02
  3. 03

Connected ideas and articles

From the conversation

Podcast episodes

Article history

Updates to this article

Tags

Free tokens or owned robots: two ways the panel would share AI's gains

On the Moonshots panel, Alex argued that China's AI loyalty perks are the start of "universal basic tokens" — redistributed access to machine intelligence — while another panelist countered that cheaper tokens will mean bigger bills, not free ones. Emad Mostaque pushed past access to ownership, proposing 100 million publicly underwritten robots owned by the people, and Alex said that sounded like communism.

6 min read

Beyond the copilot: career advice from a panel that disagrees about jobs

On Moonshots, Dave describes a hire in his early twenties who runs his agents entirely by voice, and argues the goal is to manage swarms rather than lean on a single copilot. The panel's optimistic jobs roundup runs straight into Emad Mostaque's warning that today's hiring is "the turkey before Thanksgiving" and Alex's view that every profession, trades included, is only a question of sequencing.

6 min read

Mostaque wants every child to own a slice of their state's AI company

On the Moonshots panel, Emad Mostaque presented what he calls a Champion: a locally owned intelligence utility, sold to residents at a symbolic $1 pre-money valuation before outside investors arrive, with 10% of the equity set aside in perpetuity for every child under 20. He borrows the structure from his account of how TSMC was capitalized, and argues that as the cost of intelligence falls, the money will sit in robots, deployment engineers and citizen agents. He calls it an idea, not an offering — and it leaves governance, dilution and distribution unsettled.

7 min read

The weak point in Friedberg's AI jobs optimism: workers have to choose to move

On The Diary of a CEO, investor David Friedberg argued that AI grows companies rather than shrinking payrolls, using a painter commanding five robots as his image. Pressed on entry-level hiring, Klarna's staffing numbers and a Stanford payroll study, he named the assumption he thinks could be wrong: that displaced people will see an opportunity and take it.

10 min read