It sounds like a thought experiment. But it’s actually one of the most important economic questions of our time — and almost nobody is asking it seriously.

The AI industry is growing at a pace unlike anything in corporate history. OpenAI’s annualized revenue surpassed $20 billion in 2025, up from $6 billion in 2024 — confirmed by OpenAI CFO Sarah Friar in an official blog post. 

Meanwhile, AI companies, consultants, and tech CEOs keep telling us that automation will displace millions of jobs. According to the World Economic Forum’s Future of Jobs Report 2025, 41% of employers globally plan to reduce their workforce as AI automates certain tasks. 

Here’s the paradox nobody wants to name out loud: if AI eliminates the jobs of the people who pay for AI tools, who keeps the lights on?

This isn’t sci-fi. It’s basic economics. Consumer spending — on software, subscriptions, digital tools, and services — depends on people having income. And income, for most of the world, still comes from work.

In this article, we break down who actually funds the AI boom today, what happens to that funding model as automation accelerates, and what it means for digital entrepreneurs, affiliate marketers, and anyone building a business in the age of AI.

The Paradox at the Heart of the AI Economy 

The AI industry is built on a quietly uncomfortable assumption: that the economy generating demand for AI tools will remain intact as AI reshapes that same economy.

Here’s what the numbers actually show. As of October 2025, only about 5% of ChatGPT’s weekly active users are paying subscribers — meaning the overwhelming majority of usage is subsidized by a small minority of paying customers. 

This creates a structural paradox. The businesses cutting headcount through automation are, simultaneously, the primary customers keeping AI companies profitable. Enterprise services were OpenAI’s fastest-growing segment in 2025, with over 1 million companies paying for enterprise-grade AI products. 

In other words: the payer and the disruptor are the same entity.

The WEF’s Future of Jobs Report 2025 projects that 92 million roles will be displaced while 170 million new ones will be created by 2030 — a net gain of 78 million jobs on paper, but one that masks enormous short-term disruption concentrated in specific job categories. 

The workers in those displaced categories are also consumers — of software, subscriptions, and digital services. This is not a temporary imbalance. It reflects how the AI economy is actually structured. And it raises a question that gets more urgent as automation accelerates: if enterprise keeps shrinking its workforce to fund AI efficiency gains, at what point does consumer purchasing power — for everything, not just AI — begin to erode? That’s not a philosophical question. It’s an accounting one.

Who Actually Pays for AI Services Today?

To understand the paradox, you first need to follow the money. And the money is not coming from individual users paying $20 a month.

According to Menlo Ventures’ 2025 State of Consumer AI report, nearly 1.8 billion people use AI worldwide — but only around 3% pay for premium services. Even ChatGPT, with its first-mover advantage, converts just 5% of weekly active users into paying subscribers. The remaining 95% use it for free. That means the consumer side of the AI economy is, at its core, a massive subsidized product.

The real money flows from a different direction entirely. Enterprise generative AI spending reached $13.8 billion in 2024 — a 6x increase from $2.3 billion in 2023. And corporate AI investment overall hit $252.3 billion in 2024, with private investment climbing 44.5% year-over-year. According to Andreessen Horowitz’s 2025 survey of 100 enterprise CIOs, enterprise AI budgets grew beyond already high forecasts and graduated from pilot programs into recurring line-items in core IT budgets — with leaders expecting an average of 75% further growth in the coming year.

At the organization level, the numbers are equally striking. CloudZero’s research shows the average enterprise spent $62,964 per month on AI in 2024, rising to $85,521 in 2025 — a 36% increase. The proportion of organizations spending over $100,000 per month more than doubled, from 20% in 2024 to 45% in 2025.

The picture becomes even clearer when you look at macroeconomic data. According to Fast Company’s analysis citing J.P. Morgan, AI spending contributed 1.05% of total U.S. economic growth in the first half of 2025 — up from just 0.02–0.03% between 2022 and 2024. Meanwhile, consumer spending’s contribution to GDP growth fell from 2.6% to 1.05% in the same period.

The conclusion is hard to argue with: the AI economy runs on enterprise budgets, not consumer wallets. Individual users are the product’s audience. Corporations are its customers. And those corporations are investing in AI precisely to reduce their dependence on human labor — the same labor that generates the consumer income the broader economy depends on.

How Many Jobs Is AI Actually Eliminating Right Now?

The short answer: fewer than the headlines suggest — but in ways that matter more than the raw numbers show.

According to Challenger, Gray & Christmas, nearly 55,000 U.S. job cuts were directly attributed to AI in 2025 — out of a total 1.17 million layoffs that year, the highest level since the 2020 pandemic. That’s about 4.5% of all job losses with AI explicitly cited as the reason. Among the companies that publicly named AI as a cause: Amazon eliminated 14,000 corporate roles, Microsoft cut around 15,000 jobs, and Salesforce reduced its customer support workforce by 4,000 — with CEO Marc Benioff stating AI now handles up to half of the company’s work.

But the headline layoff numbers understate the real shift. According to WEF’s Future of Jobs Report 2025, 91% of companies report that roles have already been changed or eliminated by automation — even when layoffs look small on paper. And research tracking AI-exposed job categories shows a 13% decline in employment for workers aged 22–25 in AI-exposed roles since 2022 — a pipeline shock that hits new graduates hardest and rarely makes breaking news.

The Klarna story is the clearest real-world case study.

In 2022, Klarna laid off approximately 700 employees — roughly 10% of its workforce. By early 2024, its AI assistant, built in partnership with OpenAI, was handling 2.3 million customer service conversations across 23 markets and 35 languages — about 66% of all support chats. CEO Sebastian Siemiatkowski declared it was doing the equivalent work of 700 full-time agents, resolving issues in 2 minutes versus 11 minutes for human agents, and projected $40 million in annual profit improvement.

Then reality hit. By early 2025, customer satisfaction had fallen sharply, complaints increased, and Siemiatkowski was forced to publicly admit: “Cost unfortunately seems to have been a too predominant evaluation factor.” Klarna began rehiring — this time in a gig-style model targeting students and remote workers. Investor Chamath Palihapitiya called it a warning for the entire tech sector, writing that replacing humans with “probabilistic code” is “fraught with edge cases.”

Anthropic CEO Dario Amodei has made an even bolder prediction: AI could eliminate half of all entry-level white-collar jobs within five years. Nvidia CEO Jensen Huang pushed back at VivaTech 2025, arguing that greater productivity typically leads to more hiring, not less. The honest answer is that both are partially right — and the gap between their positions is where most workers currently live.

What the data actually shows is not mass extinction of jobs, but mass restructuring — concentrated at the entry level, moving faster than retraining programs can respond, and disproportionately affecting younger workers and women, 79% of whom in the U.S. work in jobs at high risk of automation compared to 58% of men.

The jobs aren’t all disappearing. They’re changing shape — and not everyone can change with them at the same speed.

Three Economic Scenarios — and Who Pays in Each One

There is no single answer to who pays for AI when automation scales. The answer depends entirely on which economic path we actually take — and right now, all three are running in parallel.

Scenario 1: Augmentation — Humans and AI Work Together

This is the most optimistic scenario, and currently the most common narrative among tech leaders. The idea: AI makes workers more productive, companies grow faster, revenues rise, and everyone — including workers — benefits from the expanded economic pie.

The data has some real support here. According to the Federal Reserve Bank of St. Louis, workers using generative AI save an average of 5.4% of their working hours — about 2.2 hours per week for a full-time employee. A landmark Stanford and MIT study of 5,000 customer service agents found that AI assistance boosted productivity by 14%, with the biggest gains going to the least experienced workers. And new research from MIT Sloan argues that many critical human-intensive tasks have actually increased in frequency between 2016 and 2024 — meaning AI may be creating more high-skill human work, not less.

Companies that successfully adopt this model are seeing measurable results. BCG research of 1,250 companies found that “future-built” organizations — those building AI-augmented workforces rather than simply cutting headcount — generate 1.7 times more revenue growth than competitors, with anticipated revenue increases of 14.2% and cost reductions of 13.6% by 2028.

But there’s a critical caveat. According to California Management Review’s 2025 meta-analysis, a pooled analysis of 371 studies finds no robust relationship between AI adoption and aggregate labor-market gains once methodological bias is controlled. Individual productivity improves; economy-wide, the effects are far murkier. And BCG notes only 5% of companies are achieving AI value at scale — 60% report minimal returns despite substantial investment.

Who pays for AI in this scenario? Productive, employed workers and the businesses they work for. The consumer base remains intact. This is the scenario the AI industry is betting on.

Scenario 2: The UBI Model — Governments Redistribute AI Gains

If displacement outpaces augmentation, the second scenario becomes relevant: governments step in to maintain consumer purchasing power by redistributing AI productivity gains through universal or guaranteed basic income.

This is no longer a purely theoretical discussion. According to the Stanford Basic Income Lab, more than 160 UBI tests or pilots have been conducted over the past four decades. The results are nuanced but broadly encouraging. Finland’s pilot found that recipients experienced significantly better mental wellbeing and greater trust in institutions. In the Stockton, California pilot, guaranteed income recipients actually increased their full-time employment rates compared to non-recipients — challenging the assumption that cash transfers reduce the incentive to work.

The most notable experiment in the AI context was Sam Altman’s OpenAI-backed UBI study, which provided $1,000 per month to 1,000 low-income participants in Texas and Illinois for three years. The 2025 NBER evaluation found surprising results: recipients prioritized paying down debt over increased spending — improving their long-term financial stability, but not immediately stimulating consumer demand in the way advocates predicted.

AI pioneer Geoffrey Hinton, widely known as the “godfather of deep learning,” has come out in support of UBI as a response to AI-driven job displacement, stating he advised the UK government in Downing Street that UBI was a good idea. Philosopher and UBI advocate Scott Santens frames the core issue sharply: “Why should only one or two companies get rich off of the capital, the human work, that we all created?”

The challenge is fiscal scale. A program providing $1,000 per month to every American adult would cost approximately $3 trillion annually — roughly equivalent to the entire federal tax revenue in some fiscal years. As of 2025, no country has implemented a full UBI system. The politics remain far harder than the economics.

Who pays for AI in this scenario? Governments — funded by taxes on AI-driven corporate profits, wealth taxes, or “robot taxes” on automation. The consumer base is artificially maintained through redistribution. Whether this creates a virtuous cycle or a permanent dependency is the unresolved question.

Scenario 3: Winner-Takes-All — A Small Elite Funds Everything

The third scenario is already partially underway — and it’s the one least discussed in polite company.

In 2025, at least 20 billionaires gained nearly $500 billion in additional wealth from AI investments alone. AI captured close to 50% of all global venture capital in 2025, up from 34% in 2024, with foundation model companies like OpenAI and Anthropic alone capturing 14% of all global venture investment. The “Magnificent Seven” tech companies collectively accounted for nearly 35% of the entire S&P 500’s market value by late 2025, with Big Tech committing $405 billion in capital expenditure toward AI infrastructure — a moat so wide that new entrants find it nearly impossible to compete.

This concentration is not incidental. As RBC Wealth Management’s analysis shows, the eight largest Big Tech companies held $490 billion in cash and generated nearly $400 billion in free cash flow in 2025 — meaning AI expansion is being funded almost entirely from internal corporate cash, not external capital markets. A self-sustaining loop is forming: Big Tech uses AI profits to buy more compute, build better models, and capture more market share, without needing ordinary consumers to participate at all.

Brookings Institution researchers warned as early as 2021 that “winner-take-most” dynamics in AI are already shaping economic geography into a highly concentrated “superstar” system — with R&D and value creation entrenching in a handful of locations, and a growing “geography of discontent” everywhere else.

The most blunt real-world example: Forbes reported that new AI billionaires in 2025 are predominantly founders of companies built explicitly to replace human customer service, manufacturing, and administrative roles with AI software. The business model is not “make workers more productive.” It is “eliminate the workers.” At CES 2026, AI startup Sierra — backed by former Facebook and Google executives — showcased conversational AI agents designed to replace human customer service representatives for companies including Rivian and The North Face.

Who pays for AI in this scenario? Primarily the corporations and billionaires who own the infrastructure. The consumer layer becomes increasingly irrelevant to AI’s revenue model — until the broader economic contraction makes it impossible to ignore.

What History Tells Us About Technology and Purchasing Power

Every time a major technology has threatened mass unemployment, the same argument has emerged: “This time it’s different.” And every time — so far — the argument has been wrong. But the economists who study this question most carefully are increasingly reluctant to say the same about AI.

The pattern history actually shows

The historical record on technology and jobs is more complicated than either optimists or pessimists usually admit. According to IMF Finance & Development’s analysis of the Industrial Revolution, during the early decades of Britain’s industrialization, real wages stagnated even as output per worker rose. Wages only began to rise in line with productivity after the middle of the 19th century — meaning the adjustment took generations, not years. The “displacement effect” dominated first; the “reinstatement effect” — where new industries absorbed displaced workers — came much later.

Research from the Chicago Fed tracking U.S. Census data across the Second Industrial Revolution found that automation created fewer middle-skill jobs than it made obsolete — producing a “hollowing out” of the skill distribution in manufacturing. Crucially, younger workers led the transition into growing new occupations, while older workers were far more likely to remain stuck in declining roles or shift to unskilled labor. Sound familiar?

The internet followed the same script. The dot-com era promised to eliminate entire categories of work — travel agents, bookstore employees, bank tellers. Many of those jobs did shrink. But the internet also created entirely new job categories that hadn’t existed before: SEO specialists, social media managers, UX designers, growth hackers. According to research cited by the Congressional Research Service, about 60% of workers in 2018 held occupations that did not exist in 1940. The long-run net effect was positive — eventually.

So why are economists nervous this time?

Because AI breaks the core assumption that made the historical pattern work.

As Anton Korinek, economist at UVA Darden and NBER researcher, explains: “Economists have spent decades explaining why technological automation doesn’t lead to permanent unemployment. With AI, we are potentially facing something fundamentally different: a technology that could substitute for human cognitive capabilities across all domains.” He adds plainly: “This is not about falling for old fallacies — it’s about recognizing that transformative AI could change the very foundations of our economic system.”

Previous waves of automation — textile looms, factory robots, spreadsheet software — replaced specific categories of tasks. They displaced workers with particular skills while leaving untouched the vast domain of human cognitive and creative work. AI doesn’t respect that boundary. As Foreign Affairs noted in its analysis of the AI economic revolution, earlier digital technologies “had little effect on knowledge industries and creative industries such as medicine, law, advertising, and consulting.” AI has shattered that constraint.

The productivity paradox, version 2.0

There’s another historical parallel worth taking seriously — one that cuts against AI optimism in the short term.

In 1987, Nobel laureate Robert Solow made a famous observation: “You can see the computer age everywhere but in the productivity statistics.” For years after computers became widespread in offices, productivity growth actually slowed. The phenomenon became known as Solow’s productivity paradox — the gap between a technology’s theoretical potential and its measurable economic impact.

According to a February 2026 Fortune analysis, we may be living through the same paradox again. A study of 6,000 CEOs across the U.S., U.K., Germany, and Australia found the vast majority report little to no impact from AI on their operations — despite 374 S&P 500 companies mentioning AI positively on earnings calls in 2024–2025. Nobel laureate Daron Acemoglu’s own 2024 MIT research found only a 0.5% productivity increase over a decade from AI — leading him to say: “It’s just disappointing relative to the promises that people in the industry and in tech journalism are making.”

The historical lesson, then, is not “don’t worry, it always works out.” It is: technology transitions always work out eventually — but the transition itself causes real damage, concentrated in specific groups, over timeframes that feel very long if you’re the one living through them. And the people most affected are precisely the ones whose income keeps the consumer economy — and the AI subscription economy — running.

The Self-Defeating Business Model Problem

In 1914, Henry Ford made a decision that confused his competitors and shocked Wall Street. He doubled his workers’ wages to $5 a day — at a time when the average factory wage was $2.25. Ford’s primary motivation was to reduce catastrophic worker turnover — his factories had to hire over 40,000 new workers a year just to keep 13,000 on the job. But the side effect became one of the most important economic lessons of the 20th century: workers who earned enough could become customers. Ford himself later wrote, “We increased the buying power of our own people, and they increased the buying power of other people, and so on and on.”

The AI industry in 2025 is running the opposite experiment — and nobody seems to want to name what happens at the end of it.

The unit economics don’t work — and everyone knows it

The financial reality of the AI boom is stranger than most people realize.According to reporting compiled by Where’s Your Ed At, running OpenAI cost approximately $9 billion in 2024 against roughly $4 billion in revenue — meaning the company spent $2.25 for every dollar it earned. OpenAI CEO Sam Altman himself admitted that even their premium $200/month ChatGPT Pro subscription loses money because, as he put it, “people use it much more than we expected.” Anthropic lost an estimated $5.3 billion in 2024 despite impressive revenue growth. xAI reportedly burned through $1 billion per month while generating minimal revenue.

These are not scrappy startups. These are the most-funded companies in corporate history, and they are structurally unprofitable. The business model, at its core, is: raise capital, subsidize usage, hope scale eventually produces unit economics that justify the valuation. It is the same logic that powered Uber and WeWork — and it has the same unresolved question at its center: who pays when the subsidies run out?

The capital recycling loop

Making the picture stranger still, much of the AI industry’s apparent revenue is circulating within a closed loop of mutual investment. Yale Insights’ analysis of the AI bubble maps the web of cross-ownership clearly: OpenAI is committed to investing $300 billion in computing power with Oracle over five years; Nvidia is investing $100 billion in OpenAI; Microsoft — which holds equity in OpenAI — is also a major customer of CoreWeave, in which Nvidia holds a stake; and Microsoft accounted for nearly 20% of Nvidia’s revenue. As Yale’s researchers note: “You cannot help but ask, ‘Is this like the Wild West, where anything goes to get the deal done?’”

As one analyst writing for Medium put it, this creates the illusion of explosive revenue growth in each company’s financials — Oracle books billions from OpenAI, Nvidia books record GPU sales — while the ultimate source of funds is their mutual investments. Nobody outside the loop is generating new demand. The money is recycling, getting burned at every step.

The demand problem nobody models

And here is where the Henry Ford logic reasserts itself — in reverse.

A 2025 study published in Scientific Reports, modelling AI’s impact on the Australian economy, found that a moderate pace of AI adoption without corrective redistributive policies could lead to significantly lower consumer spending by mid-century. The mechanism is straightforward: AI-driven productivity gains that are not passed on to workers as wages or to consumers as prices concentrate wealth at the top while suppressing the consumer demand that drives the broader economy — including demand for software, subscriptions, and digital tools.

The AI valuations currently propping up markets are, as Yale’s researchers note, predicated on massive future revenues. Those revenues assume a thriving market for AI products and services. That market assumes people have income to spend. And income, for most people, still comes from the jobs AI is being deployed to eliminate.

As the Bulletin of the Atomic Scientists reported in December 2025, AI-related stocks had already accounted for 75% of S&P 500 returns since ChatGPT’s launch. The top 20 companies on the S&P 500 now account for 52% of total market value — with AI-related companies dominating the top 10. A deflation of the AI bubble on the scale of the dot-com collapse, former IMF chief economist Gita Gopinath has warned, could have “severe global consequences.”

There is a word for what happens when an industry eliminates the purchasing power of its own future customers while simultaneously depending on unsustainable capital subsidies to stay afloat. Economic historian Richard Murphy, writing in September 2025, framed it as the paradox Ford understood and the AI industry is ignoring: “Labour is both a cost in the production ledger and the foundation of demand in the wider economy. Ignore the second role, and you collapse the market you depend upon.”

Ford understood that his workers were not just costs. They were customers. The AI industry, for now, is betting it never has to learn the same lesson.

What This Means for Affiliate Marketers and Digital Entrepreneurs

The economic questions explored in this article aren’t abstract for performance marketers. They land directly on your business model — in ways that are already visible today, and ways that are about to get much harder to ignore.

The good news first: the AI affiliate window is open and genuinely lucrative

According to Affiverse’s 2025 guide to AI affiliate programs, most AI affiliate programs currently offer 30% to 50% commission, with many providing recurring revenue on renewals — commission levels that are rare in most mature affiliate verticals. Unlike traditional programs paying single-digit percentages on physical goods, AI SaaS products have high profit margins and strong customer lifetime values that justify generous structures. Post Affiliate Pro’s research confirms that AI and machine learning SaaS companies are currently averaging 24.5% commissions — the highest of any software category — precisely because competition for quality affiliates is intense.

For affiliates focused on B2B tools, productivity software, and AI-native platforms, this is a meaningful window. As Affiverse notes, affiliate marketing is also solving a core monetization challenge for AI companies themselves: instead of betting everything on direct sales, AI companies can leverage affiliate networks to distribute risk across thousands of marketers — which is particularly valuable when user behavior is unpredictable and acquisition costs remain high. In practical terms, you’re solving a real problem for the industry, not just clipping commissions.

The structural threat: agentic commerce is rewriting the purchase journey

The more urgent issue for affiliates is not whether AI replaces their audience’s jobs — it is whether AI replaces their role in the purchase journey entirely.

In September 2025, OpenAI launched Instant Checkout in ChatGPT, built on the Agentic Commerce Protocol co-developed with Stripe. The mechanism is straightforward: a user asks ChatGPT for product recommendations, sees results, and completes a purchase without ever leaving the chat interface. More than 700 million people turn to ChatGPT each week, and the initial rollout gave over one million Shopify merchants — including Glossier, SKIMS, Spanx, and Vuori — the ability to accept purchases directly within the conversation. By October 2025, PayPal had adopted ACP to power in-chat payments within ChatGPT, extending agentic commerce to PayPal’s global merchant network. And in January 2026, Google launched its own Universal Commerce Protocol with partners including Shopify, Etsy, Wayfair, Target, and Walmart — signaling that AI-native checkout is becoming the industry standard, not an experiment.

The attribution problem this creates is significant. As Search Engine Land reported in November 2025, zero-click searches now make up nearly 60% of Google’s mobile queries, and sites previously ranked first can lose up to 79% of their traffic when pushed below an AI Overview. For affiliates whose entire revenue model depends on tracked clicks and attributable conversions, a purchase completed inside ChatGPT — based on a recommendation informed by your content — may generate zero commission. The click never happened.

As eMarketer’s senior analyst Max Willens noted in an October 2025 briefing, AI-driven discovery is making attribution “even murkier” — but he was also direct that AI won’t put affiliates out of business: the relationship between affiliate content and AI recommendations is “the kind of relationship that will grow symbiotically.” The challenge is capturing value from that relationship before measurement infrastructure catches up.

MarTech’s analysis of the shift frames the core issue clearly: traditional last-click attribution no longer tells the full story. AI-powered discovery pushes influence to the earliest stages of the customer journey — long before measurable engagement begins. Affiliates who use clear UTM parameters, API-based integrations, and real-time tracking gain a distinct edge because these systems can map activity across platforms and reveal AI-driven discovery’s true contribution to conversion.

The audience displacement risk — the slow-moving threat

There is a second, slower-moving threat that connects directly to the macro question in this article. The buyers your affiliate content targets today are disproportionately employed middle-class consumers — the same demographic that WEF’s research identifies as most exposed to AI-driven job displacement at the entry and mid-skill levels. A contraction in that audience’s purchasing power is not a distant scenario. It is a demand-side risk with a specific demographic profile, arriving gradually enough to miss until it’s already underway.

The practical response

The affiliates and digital entrepreneurs best positioned in this environment share one characteristic above all others. According to Retail TouchPoints’ October 2025 analysis of agentic commerce, the affiliate model is not being disrupted by agentic commerce — it is being absorbed into it. AI agents are, in effect, becoming the new publishers: they discover, compare, and recommend products on behalf of consumers. The commission logic remains intact. What changes is the infrastructure required to participate in it.

Practically, this means three things right now. First, treat your product feeds, metadata, and structured data as the new SEO — AI agents rank merchants based on data quality, not page authority. Second, invest in expert-led, first-hand reviews and comparisons: as MarTech notes, AI models increasingly prioritize trustworthy, original, experience-based content — the kind that affiliate publishers are best positioned to produce and that pure AI content cannot manufacture at scale. Third, diversify your attribution infrastructure now, before the measurement gap between AI-influenced and AI-attributed conversions becomes your primary revenue problem.

The window for that adaptation is open. It is not indefinite.

Conclusion: The Real Question Isn’t Technical

The question in this article’s title — who will pay for AI services if AI replaces everyone? — sounds provocative. But it’s not a rhetorical trick. It’s a genuine stress test of the economic model underlying the most consequential technology deployment in modern history.

Here’s what the evidence actually shows. Right now, the AI economy is being funded primarily by enterprise budgets and institutional capital, not by individual consumers. The consumer base — the workers who pay for subscriptions, tools, and digital services — is structurally marginal to the AI revenue model today. That’s the first uncomfortable fact. The second is that the same corporations funding AI are deploying it specifically to reduce their dependence on labor. And the third is that this creates a loop with no clean exit: eliminate purchasing power, and you eventually eliminate the market you depend upon — including for AI.

As Anthropic’s own economic policy research published in October 2025 acknowledges, the company’s own user data shows people are rapidly shifting from “collaborating” with AI to “delegating” entire tasks to it — and the workforce implications are uncertain enough that the company has begun working with economists to map out policy responses across three scenarios, from mild disruption to dramatic job losses. Anthropic’s researchers and their Economic Advisory Council are exploring everything from automation taxes to sovereign wealth funds that would give citizens direct stakes in AI revenues. These are not fringe ideas — they are being modeled by the same institutions building the technology.

The IMF’s April 2025 working paper on AI and inequality found that AI’s impact on wealth distribution is more complex than either optimists or pessimists claim. Unlike previous waves of automation that increased both wage and wealth inequality, AI could reduce wage inequality by disrupting high-income cognitive tasks — while simultaneously increasing wealth inequality through higher capital returns. In other words: the middle class may shrink more slowly than feared, while the gap between capital owners and everyone else widens faster than expected. The Gini coefficient in the IMF’s models rises by 2 to 7 percentage points depending on the pace of adoption and policy response.

Research from NBER economists Anton Korinek and colleagues makes the distributional logic plain: when AI substitutes for human labor rather than complementing it, it generates a zero-sum redistribution from labor to capital in the short run. The workers who lose out would rationally oppose the innovation — and if innovators wish to maintain their position, “it would behoove them to think harder about how to engage in redistribution.” Taxing the winners in winner-takes-all outcomes, the research suggests, may have far less negative effect on innovation incentives than is commonly assumed.

The real question, then, is not whether AI can replace human workers. Technically, it can replace more and more of them. The real question is who controls the economic output that AI generates — and how that output is distributed back into the economy that AI itself depends on.

There are only a few possible answers. Either corporations voluntarily share productivity gains with workers and consumers — which history suggests is unlikely without pressure. Or governments intervene to redistribute AI gains through taxation, sovereign wealth funds, or income support — which is politically hard but increasingly discussed at the highest levels of policy. Or the concentration of AI wealth continues until consumer purchasing power contracts enough to trigger a correction — a self-defeating cycle that Ford understood a century ago and that the AI industry is, for now, choosing to ignore.

The technology is not the hard part. Building a transformer architecture, training a large language model, deploying an AI agent that handles customer service — these are engineering problems, and they are being solved. The hard part is what has always been the hard part in every industrial revolution: deciding who benefits, and making sure enough people benefit that the system doesn’t collapse under the weight of its own inequality.

That is not a technical question. It is a political one. And the window for answering it thoughtfully is shorter than most people realize.

FAQ