Anthropic's $44T GDP Scenario Comes With 12% Unemployment

 

By Maya Krishnan | 18 October 2026

Photo: Thai national flags flying outside the Queen Sirikit National Convention Center in Bangkok, venue of the 2026 IMF–World Bank Annual Meetings (file photo, July 2023). Credit: Jonashtand / Wikimedia Commons (CC BY-SA 4.0)

The biggest number to emerge from the IMF's Annual Meetings this week did not come from the Fund's own economists. It came from a tech executive on a Saturday lecture stage in Bangkok: $44 trillion — the size Anthropic co-founder Jack Clark said the American economy could reach by 2030 if artificial intelligence keeps accelerating. Clark called it, in his own words, "sounds very outlandish, I know" — and then he spent the rest of the hour defending it anyway.

Here is the part almost nobody who repeated the number bothered to print. The $44 trillion future is also the ugliest one in Anthropic's own model. It arrives with unemployment at 11.9 per cent — worse than anything the United States has seen since the Second World War — knowledge-worker wages down more than a tenth, and labour's share of national income collapsing from about 60 per cent to 45.2. This is not the good-news version of AI. It is the one in which the economy grows and the workers don't — and Anthropic itself assigns it no probability at all.

What Clark told the IMF's policymakers

Clark spoke on Saturday, 17 October, in a lecture to hundreds of government and finance officials at the Queen Sirikit National Convention Center, where the IMF and World Bank are holding their 2026 Annual Meetings from 12 to 18 October. Reuters' David Lawder, reporting from Bangkok, carried the $44 trillion figure first; Channel News Asia's longer version of the remarks added the warnings that Reuters trimmed.

"We should expect that in the coming years, the whole pace of progress of not just AI but science as a whole will speed up, and this will feel confusing and disorienting, and will bring with it great wealth and potentially great peril," he said in a lecture to hundreds of government and finance officials, according to CNA. On the dangers of AI-assisted science, he added: "Being able to use AI for science opens us up to the misuses of science as well, biological and cyber weapons and things we cannot yet imagine, and we have many choices to make."

The policy prescription, in Clark's telling, is "unusual and serious rules" — international cooperation of a kind rarely seen outside arms control — and the window to impose safeguards was, he said, now "a small number of years." (That phrase is Clark's; the "shrinking window for policymakers to act" framing around it is the reporters'.) CNA also reported growing evidence, in Clark's account, of a "singularity" where AI development outpaces human control — that one is the outlet's paraphrase, not a direct quote, and worth flagging as such.

Clark, a former journalist who now serves as Anthropic's head of policy, knows exactly how a headline travels — which makes the order of his caveats telling. He led with "sounds very outlandish, I know," laughing through the implausibility, and only then argued the figure was implied by the technology's trajectory. When the pitchman opens with the disclaimer, the disclaimer is part of the pitch.

Three scenarios, one of them ugly

The $44 trillion figure is not a forecast. It is one prong of a three-scenario model published on 9 September 2026 by Anthropic's Economics team — Scenarios for our Economic Future, accompanied by an interactive Econ Scenario Explorer — and AI Weekly notes that "Anthropic assigns no probabilities to any scenario, making the tool a structured disclosure without a forecast or implied likelihood." All figures are in 2025 prices.

The modest scenario puts US output at $34.1 trillion in 2030, just 1.6 per cent above a no-AI baseline — internet-scale change, roughly 2.4 per cent annual growth. The substantial scenario reaches $36.3 trillion, 8.3 per cent above baseline: AI doing about half of knowledge work by 2030, mostly autonomously, with knowledge-worker wages stagnating. And the extreme scenario: $44.4 trillion, 32.4 per cent above baseline, which requires recursive self-improving AI and high autonomy — annual growth of around 15 per cent in the extreme years.

The technical report, Economic Scenarios for Transformative AI, was authored by Anton Korinek and colleagues and externally reviewed by Daron Acemoglu, David Autor, Ben Jones and Pete Klenow — serious names, which is precisely why the lack of assigned probabilities matters. A companion Morning Consult survey of 10,980 American adults in August 2026 found the typical respondent's answers implied the substantial path — about 10 per cent higher GDP by 2030, around 5 per cent unemployment — with only about one in ten landing near the extreme.

Why $44 trillion comes with 12 per cent unemployment

This is the section the weekend coverage skipped. In the extreme scenario, "cognitive unemployment" — white-collar work evaporated — hits 17.9 per cent, and overall unemployment reaches 11.9 per cent, worse than anything since the war. Knowledge-worker wages fall more than a tenth, and labour's share of income drops from roughly 60 per cent to 45.2 per cent.

We checked Anthropic's public Econ Scenario Explorer ourselves — the $44.4 trillion figure sits inside a slider-adjusted dashboard, not a prediction, and the tool offers no confidence level for any path. The pack treated a company-built scenario tool as a headline number; the company's own documentation treats it as a structured thought experiment. That is not a small distinction. It is the distinction.

Clark was, to be fair, candid about the labour side. He said AI was already destroying jobs in fields with "high AI exposure" — he named call centres and customer service personnel — while enhancing employment in scientific research, and called job creation from new industries "a major unknown." The honest translation: the model knows how to count the gains and has no idea how to count the jobs.

Here is my own judgment, and I will not dress it up as neutrality: the $44 trillion number is a disclosure disguised as a boast. Anthropic built a scenario tool that says, in its own numbers, that the extreme AI future is a jobs catastrophe — then sent its co-founder to the IMF to lead with the GDP figure and bury the unemployment rate. The honest headline was always the labour one. The company chose the money one. Draw your own conclusions about why.

What would have to be true for $44 trillion

Do the arithmetic — we did. Going from just over $30 trillion to $44.4 trillion in four years means roughly 48 per cent cumulative growth, or about 10.3 per cent a year, in real terms. The United States has never sustained anything like that outside wartime. The postwar record is roughly 7 per cent in a single year and 4 to 5 per cent across a full boom. Clark's extreme path runs at four to five times the pace of the IMF's own current forecast for the American economy — 2.3 per cent growth in both 2026 and 2027. An economy does not grow at 10 per cent a year on vibes; it needs new industries, new inventions, new everything.

Clark concedes the point himself. "I know, but it's implicated by the technology continuing to get better. And if GDP were to grow this much, then economic doctrine suggests that new industries and inventions would need to have been created to power such a major economic expansion." The quote is doing a lot of heavy lifting there: the scenario's own premise is that the machines invent the growth industries faster than the economy can shed the workers. That is not a plan. It is a hope wearing a spreadsheet.

The bottlenecks are equally obvious and unquantified in the headlines: data centres, memory and chip supply, trillions in financing. Independent analysts cited by aggregators this week noted that financing a $14 trillion expansion could itself fuel inflation or financial-stability risks — the boom could eat its own foundations before 2030. None of that made the first wave of coverage, which mostly copy-pasted the $44 trillion and moved on. That was shoddy.

The same week, the IMF said the opposite

Here is the irony that belongs in every telling of this story. In the very same Annual Meetings, days earlier, the IMF's own downside scenario warned that an AI-driven market correction could cut US output by 1.8 per cent by 2028 — wiping out the Fund's expected gains — and shave 1.1 per cent off global GDP, as we reported this week. One institution, one week, two futures: AI adds $14 trillion by 2030, or AI subtracts nearly two per cent by 2028.

That is not a contradiction to resolve; it is the actual story. The honest band of uncertainty around AI's economic impact is the widest it has ever been, and both poles were on the table in Bangkok. The IMF's World Economic Outlook, which forecasts US growth of 2.3 per cent this year and next alongside Taiwan's 10.7 per cent AI boom, sits roughly in the middle of a room where the walls are $44 trillion apart. Anyone selling you certainty about any of these numbers — including Clark, and including the Fund — is selling you something else.

Consider the source: a November IPO

Reuters' original report carried a detail that deserves more airtime than it got: Clark's remarks come as Anthropic prepares for an initial public offering expected in November that could value the company at up to $2 trillion. That is Reuters' reporting, not Anthropic's announcement — but it reframes the lecture. A $44 trillion AI future is a useful story to tell finance ministers when you are about to ask public markets for a $2 trillion valuation.

As we examined earlier this week, Anthropic's 45 per cent revenue growth wasn't enough to quiet the questions around that valuation — growth that would thrill any normal company looked thin against a two-trillion-dollar price tag. Numbers like $44 trillion do not exist in a vacuum; they exist in a roadshow. Read the IMF lecture as economic analysis if you like. It is also marketing, and it would be naive to pretend otherwise.

Common-sense standards and the 2027 clock

A day earlier, on 16 October, Clark had appeared on a panel titled "AI and the Global Economy: From Risk to Opportunity" alongside IMF managing director Kristalina Georgieva and Bank of Canada governor Tiff Macklem. "We need common sense standards about how you ultimately give confidence in the technology and build the economy," he told the panel, according to Nikkei Asia — and noted that companies successfully adopting AI tend to engage in "creative destruction" of themselves.

Clark also used the week to reaffirm — not announce — his moved-up timeline for recursive self-improvement. In his personal newsletter Import AI on 12 October, he wrote: "I now think it's more likely we'll see full RSI (recursive self-improvement) become possible in 2027, whereas before I thought it was more likely in 2028." He repeated the 2027 call in an NHK interview in Tokyo the same week: "My belief now is that AI systems may be capable of doing AI research themselves in 2027." The basis he cites is the TasteVal benchmark from P-Zero Research, which finds AI's "experimental research taste" doubling roughly every three months since December 2025, plus Claude agents leading 26 per cent of Anthropic's core AI research tasks in August 2026, up from under one per cent in February. His previous position, set out in May, was roughly 30 per cent odds for 2027 and around 60 per cent by the end of 2028.

So the full picture from Bangkok: an executive who believes self-improving AI arrives next year, asking governments to write "unusual and serious rules" within "a small number of years," while his company publishes a scenario in which the economy booms and the workforce breaks — and takes the company public next month. Whether you find that bracing or alarming probably depends on whether you own the shares or hold the job.

Sources: Reuters (David Lawder, Bangkok, 17 Oct 2026); Channel News Asia; Nikkei Asia; NDTV; NHK World; Financial Express; AI Weekly; CryptoBriefing; AP via The Korea Times; Tech Times; Anthropic Institute, Scenarios for our Economic Future (9 Sept 2026).

Maya Krishnan is TechPulse USA's Senior Technology Correspondent, covering the AI economy and the money behind the machines. She has a spreadsheet of every AI valuation claim since 2023 and trusts none of them.