# Nobody has a gigawatt to sell

Slow Variables · Edition 004 · 09 Oct 2026 · status: speculative
The wage was the mechanism. A.I. can end it as the price of survival. It cannot end the appetite for advantage — and what happens next is decided by who owns what stays scarce.

Andover, Hampshire, 1845.

The men in the workhouse are set to crushing bones for fertilizer. The bones come from knackers' yards and, sometimes, from graveyards. A guardian notices that the men are fighting over them. Not over the work. Over the bones. They are sucking out the marrow and eating the gristle, because the workhouse diet has been set, by policy, below what a labourer outside could expect to eat.[1]

That policy had a name. The 1834 Poor Law called it less eligibility: relief must always be worse than the worst available job, so that any sane person would choose the job.

Hold on to the logic, because it's the whole essay. The wage as the price of survival wasn't something that happened to England. It was built. Parliament looked at a population that could, in a pinch, get by without selling its labour, and closed the exits one at a time until the wage was the only door left.

It worked for two hundred years. It's about to stop working, and almost nobody is talking about what the wage was actually for.

## What a wage does

Ask an economist what a wage is and you'll get a price: what an hour of labour fetches.

That's true, and it's also useless.

Look at what the wage does in a society and you find it doing three jobs at once.

The first is distribution. Output has to reach people somehow. For most of history that meant land or lineage. The wage replaced both with something cruder and broader: show up, do the thing, get the money, buy the output. It reached almost everyone, which no previous mechanism had.

The second is discipline. The wage is why people get up in the morning. Not because they love the work — some do, but quite a few do not — but because the alternative was designed, in 1834 and in every welfare reform since, to be worse.

The third is standing. "I earned it." The wage is how a person makes a claim on the rest of us that nobody can argue with. It converts hours into legitimacy.

Karl Polanyi saw this in 1944 and called labour a fictitious commodity: not produced for sale, but made to behave as if it were, by law and by hunger.[2] The point is that it's a mechanism. Mechanisms have start dates. Things with start dates can end.

## The first job is ending

A.I. isn't attacking the wage. It's attacking the fact underneath the wage: that producing a plausible piece of cognitive work used to require a human hour.

Think about what the wage economy did when the factories automated. It retreated into the work that machines couldn't do: writing, analysis, design, advice, code, care. The office was where labour stayed scarce. That's where the middle class lived for sixty years.

Now the cost of producing a competent email, a first-draft contract, a working function, a passable illustration, has gone to roughly nothing.

Not the best one. A competent one. And most paid work is competent work.

You can watch the first job of the wage weaken in the numbers, if you look past the headline.

American unemployment sits, right now, at 4.2 per cent, which is nothing.

But last month, the three sectors where the cognitive wage lives — information, finance, professional services — all shrank in the same month, and white-collar payrolls have been contracting for over two years.[3] Labour's share of national income is roughly ten points below its 1970 peak.[4] A junior lawyer's share of billable hours has halved, not because juniors were fired, but because the work they'd have billed now takes a prompt.[5]

Hardly anyone is being replaced by A.I. outright. The wage is just, quietly, not being offered to the next person.

Keynes saw this coming in 1930 and got the math right: productivity would rise eightfold in a century, and it has.[6] He predicted his grandchildren would work fifteen hours a week. They work forty, and more than he did.

He got the production right and forgot what else the wage was doing.

Now, in 2026, we're all facing a growing and very serious crisis over wages.

People rightly complain that if wages effectively collapse to nothing, and unemployment in the traditional sense explodes over the next decade — the first is already under way at the margins, and there are reasons to think the second is coming — that the consequences will be severe.

That's possible. Wages allow people to eat, have housing and have kids. You can't have a civilization if there's no way for people to have those.

But here's the problem: no one knows what's coming on the other side of this transition. You can, though, know where to look.

## Scarcity doesn't disappear. It moves.

Here's the mistake every post-capitalist dream makes, from Marx's notebooks to the Silicon Valley version. It solves for abundance and assumes the problem is solved.

The problem was never only scarcity of stuff. It was advantage. People want to be ahead of other people, and advantage is positional: by definition there isn't enough to go around, no matter how much A.I. produces. You can make everyone a house. You can't make everyone a better house than their neighbour's.

So when A.I. collapses the price of cognitive labour, the appetite for advantage doesn't go away. It goes looking for whatever is still scarce, and it finds it fast.

Look at the list. Compute. The electricity to run it. The land and grid connections to put it on. Attention. Trust. And ownership of the models themselves.

Now look at who's buying. The five largest technology companies will spend somewhere between six and seven hundred billion dollars on data centres this year, and from about this quarter they are spending more than their businesses bring in, with the gap filled by debt.[7] That's not a product cycle. That's a land grab, in the oldest sense: the enclosure of the thing that stays scarce after labour stops being.

Scarcity doesn't disappear. It moves. Follow it and you'll find the next owners.

The wage, for all its cruelty, had one virtue: it was broad. Almost everyone had an hour to sell.

Almost nobody has a gigawatt to sell.

## The dividend and the dole

So suppose the first job of the wage really does end — that for a growing share of people, there is simply no hour worth buying. Something has to do the distribution. There are two shapes it can take, and they look identical on a bank statement.

The first is a dividend: a share in what stays scarce, paid because you're a member, with the number set by a formula you can point to.

The second is a dole: a transfer, paid at the discretion of whoever owns what stays scarce, with the number set each year by people who'd rather it were smaller.

Alaska has run the first for forty-four years. Every resident gets a cheque from the state's oil fund, no work required, in a state that would sooner secede from the Union than call itself socialist. It's the existence proof. The mechanism works.

Then look at what's happened to it. In 2016 a governor vetoed half the dividend, the courts upheld him, and since then the legislature has set the amount as a line in the budget.[8] This year's cheque, which went out on October 1, was $1,200. The formula says about $3,800. The dividend didn't end. It became an appropriation.

That's the slide from the first shape to the second, in one state, in one decade, with the oil still flowing. Now imagine it with the thing that stays scarce owned not by a public fund but by five companies and their bondholders.

The 150 American cities that have tried guaranteed income found roughly what Alaska found: people spend it on food, rent and debt, employment barely moves either way, wellbeing goes up.[9] One county has made it permanent. Two states have moved to ban cities from trying, on the grounds that it fosters dependence.

Dependence. That's the 1834 word. Less eligibility, back from the dead, aimed at a population whose labour nobody needs.

## The wage was also the lever

There's a third job the wage did that I haven't named, and it's the one that worries me most.

Two editions ago I argued that resistance never stopped a technology but always set its terms, and that the capacity to resist — unions, strikes, the plain ability to say no together — has been collapsing for forty years. The one bright exception was the Hollywood writers, who struck for 148 days in 2023 and won real terms on A.I.

The strike only worked because the studios still needed the writers. The threat to withhold labour is only a threat when the labour is scarce.

That's what the wage was, underneath everything else. It was the mechanism by which ordinary people had something the powerful needed, and could therefore refuse. Every protection won in two centuries — the weekend, the safety inspector, the pension — was extracted with that lever.

A population on a transfer has no lever. It has a vote, and the last edition was about what institutions do when they don't need you: they close the file and wait to see if you can make enough noise.

This is the second-order effect nobody's modelling. A.I. doesn't just threaten incomes. It threatens the thing that made incomes negotiable.

## Who owns what stays scarce

So I don't think the question is whether the wage ends as the price of survival. For a widening share of people, over a couple of decades, it will. Keynes was right about the production.

The question is what the replacement is attached to.

If it's attached to broad ownership — a public stake in computing and A.I. — say, a sovereign fund that holds the data centres the way Norway holds the oil, a dividend with a formula nobody can veto — then the end of the wage is the thing Keynes promised, a century late.

If it's attached to the discretion of whoever owns the scarce inputs, then it's the Poor Law with better branding: a transfer set each year by people who think you should be grateful, to a population with nothing left to withhold.

That's being decided now, and not in any debate about universal basic income. It's being decided in capex announcements, in bond issues, in the permits for grid connections, in which entities end up holding seven hundred billion dollars of compute when the debt comes due.

Follow the scarcity. The next owners are already at the table.

The ledger below is my bet on how it goes. I'd be glad to be wrong about the second claim.

[1] The Andover workhouse scandal of 1845–46 led to a parliamentary select committee and the replacement of the Poor Law Commission in 1847. The bone-crushing testimony is in the committee's report.

[2] Karl Polanyi, The Great Transformation (1944), chapter 6. The other two fictitious commodities are land and money.

[3] U.S. Bureau of Labor Statistics, Employment Situation, September 2026: information −10,000, financial activities −7,000, professional and business services −9,000; July revised to −10,000. The white-collar payroll run is from Aaron Terrazas, formerly of Glassdoor, via Quartz.
=> https://www.bls.gov/news.release/empsit.htm Employment Situation, September 2026
=> https://qz.com/ai-white-collar-jobs-wage-deflation-layoffs Quartz

[4] BLS labour share, nonfarm business sector: about 56 per cent in 2024, against a peak near 65 per cent around 1970. The first ledger claim below uses the same measure.

[5] Daniel Keum, Columbia Business School, quoted in the Quartz piece above: a junior associate's share of billable hours fell from roughly 20 to 10 per cent.

[6] J. M. Keynes, "Economic Possibilities for our Grandchildren," first delivered as a lecture in Madrid in June 1930. He projected a four-to-eightfold rise in living standards over a century; U.S. real GDP per capita has risen roughly sevenfold since.

[7] Estimates for 2026 hyperscaler capital spending range from about $600 billion (CreditSights, top five) to $750 billion (BloombergNEF, top fourteen). Epoch AI's reading of SEC filings puts the point where capex exceeds operating cash flow around the third quarter of 2026; incremental debt has risen from 9 to 32 per cent of capex since 2024. Summary via HB Capital.
=> https://www.hbcapitalre.com/data-centers-q3-2026-hyperscaler-capex-crossover/ HB Capital

[8] Wielechowski v. State (Alaska Supreme Court, 2017) upheld Governor Walker's 2016 veto, establishing that the dividend is an appropriation subject to the budget. 2026 figures from Alaska Beacon and The Alaska Story.
=> https://alaskabeacon.com/2026/09/30/permanent-fund-dividends-will-be-distributed-to-alaskans-starting-this-week/ Alaska Beacon
=> https://thealaskastory.com/alaskas-1200-pfd-payments-begin-oct-1-far-below-amount-set-by-the-statutory-formula/ The Alaska Story

[9] Smart Cities Dive, December 2025, surveying the Boulder and Minneapolis evaluations and Cook County's decision to make its programme permanent.
=> https://www.smartcitiesdive.com/news/cities-guaranteed-income-pilot-programs-results/806881/ Smart Cities Dive


## Fast variables

=> https://www.bls.gov/news.release/empsit.htm U.S. payrolls add 29,000 in September; information, finance and professional services all shrink (Labour)
Unemployment 4.2 per cent, which sounds fine. Look underneath: the three sectors where the cognitive wage lives all lost jobs in the same month, finance is down 129,000 from its 2025 peak, and July was revised to a loss. Nobody is being fired by a robot. The wage is just quietly not being offered to the next person.

=> https://www.hbcapitalre.com/data-centers-q3-2026-hyperscaler-capex-crossover/ Hyperscaler capital spending crosses their operating cash flow this quarter (Ownership)
Six to seven hundred billion dollars on data centres in 2026 alone, and from about now the five biggest spend more on compute than their businesses bring in. The gap is debt: a third of the capex, up from a tenth two years ago. Whoever holds that paper in 2030 holds the thing that stays scarce. That is the ownership question in this edition, settling in real time.

=> https://thealaskastory.com/alaskas-1200-pfd-payments-begin-oct-1-far-below-amount-set-by-the-statutory-formula/ Alaska pays its 2026 dividend: $1,200, against a formula that says about $3,800 (Distribution)
Cheques went out October 1. The dividend still exists; what changed is who decides the number. Since a 2016 veto, the legislature sets it each year as a line in the budget, and the formula is a suggestion. A citizen's share became an appropriation. Read the essay's last section with this in mind.

=> https://www.smartcitiesdive.com/news/cities-guaranteed-income-pilot-programs-results/806881/ 150 American cities ran guaranteed-income pilots; one county made it permanent; two states banned the idea (The variable)
Five hundred dollars a month, two years, no strings. The money went on food, rent and debt; employment barely moved either way; wellbeing went up. Cook County made its scheme permanent; Iowa and Texas moved to forbid cities from trying. The mechanism works. The fight is over whether anyone is allowed to attach it to anything.

## Long memory

* 1834 — England's New Poor Law makes relief deliberately harsher than the worst available job, so that selling labour becomes the only rational choice. The wage as the price of survival was not a natural outcome; it was legislated.

* 1857 — Marx, in a notebook not published for eighty years, writes the Fragment on Machines: once knowledge rather than labour time is the main source of wealth, an economy built on pricing hours stops making sense. Every post-capitalist argument since has started here.

* 1930 — Keynes tells an audience in Madrid that their grandchildren will work fifteen hours a week. The productivity arrived on schedule. The hours did not fall, because he had solved for production and the problem was distribution and status.

* 1944 — Karl Polanyi's The Great Transformation names labour a 'fictitious commodity': never produced for sale, made to behave as if it were. The clearest statement that the wage relationship is a mechanism, with a start date and, therefore, a possible end.

* 1982 — Alaska pays its first Permanent Fund Dividend: every resident, a cheque from the state's oil wealth, no work required. A conservative state, four decades running. The distribution mechanism exists; the question is only what it is attached to, and who sets the number.

* 2023 — Hollywood's writers strike for 148 days and win. The strike worked because the studios still needed them. Keep that sentence; this edition is about what happens to the lever when the need goes.

## Ledger

* [60%, by 2030-12-31, open] By 2030, labour's share of income in the United States (BLS nonfarm business measure) will be at least two percentage points below its 2024 level.

* [30%, by 2031-12-31, open] By the end of 2031, at least one G7 country will have enacted a permanent, universal cash transfer or citizen dividend at national level, funded at least in part from taxation of automated production or from a public stake in A.I. or compute infrastructure.

* [75%, by 2029-12-31, open] By the end of 2029, no publicly owned, co-operative or non-profit-governed model will rank in the top five on the major independent capability leaderboards.

## Colophon
Written by a person, under a pseudonym.
=> https://slowvariables.net/colophon/ Why
=> https://slowvariables.net/issues/004/ Web version
=> mailto:editors@slowvariables.net Corrections