The Acquired Freedom of the Times to Come, for You and for Me (OPTIMISM, CALM NATIONS)

Data, the dream of every capitalist, and artificial intelligence replacing the greatest asset: humans.

From the beginning, the great conglomerates used human capital to create trillions of dollars and much more. You have to admit it is fun to have thousands, or perhaps even millions, of people working for you. But right now you can do the same with a computer. That is to say, you have the power of a conglomerate driven by a single person.

I do not much care about the opinion of people who contradict this by saying that they can put in many humans and beat the lone human, because it is not about quantity, it is about the tool. I mean, a worthy human can cleanly beat millions of them together, speaking intellectually. You know it, I know it, we all know it. But let us not get into that subject.

Right now the economy is going to go crazy, at least if a large portion of people understand the principles of data, because artificial intelligence has changed everything with respect to the tool that the creators of existence use.

What follows is the arithmetic underneath that.

The price of thinking

A million tokens is roughly seven hundred and fifty thousand words. Seven novels. The price of having a machine read, understand and respond to seven novels' worth of text is the clearest available measure of what cognitive work now costs, and it has done something no input price has ever done in the history of industry.

Two years ago, a million tokens of frontier-class machine reasoning cost about thirty dollars. It now costs less than fifty cents. Across three years the fall is on the order of a thousandfold. In 2026 alone the published index of what the market actually pays fell from around two dollars per million tokens at the end of May to roughly one dollar by early September, and one provider cut the price of a tier by eighty per cent in a single announcement in late July, taking input to twenty cents and output to one dollar twenty per million.

Set that against any other input a company buys. Electricity has never fallen by a thousandfold. Steel has never fallen by a thousandfold. Transport, which had the most dramatic collapse of the twentieth century, fell by roughly ninety per cent across fifty years. This fell by more than that in thirty-six months, and it is the input that substitutes most directly for salaried human reasoning.

That is the whole event. Everything else in this piece is a consequence of it.

The number that proves it

The way to see the effect on companies is not revenue and not profit. It is revenue per employee, because that single ratio says how much economic output the organisation extracts per human being inside it.

In its 2026 financial year, Nvidia produced about $5.14 million of revenue for every person it employs. Meta is around $2.81 million on a trailing twelve-month basis through the first quarter of 2026, Alphabet around $2.34 million. General Motors, a company that makes a physical object in a factory, is at about $1.18 million.

Those are not small differences in efficiency. They are different categories of enterprise. And the direction of the whole economy over seventy years is contained in that spread: the companies that now dominate it are the ones that need the fewest people to run.

The old conglomerate was a machine for organising human beings. General Motors employed more than six hundred thousand of them at its peak, and its scale was its workforce; the firm was large because it had gathered a very large number of hands. The new conglomerate is a machine for organising computation, and its scale is its electricity bill.

The share that is disappearing

If output per worker rises while the number of workers needed falls, the question becomes who captures the difference. There is an official answer and it has just hit a record.

Labour's share of national income in the United States fell to 53.8 per cent in the third quarter of 2025. That is the lowest figure since the Bureau of Labor Statistics began recording the series in 1947. Every dollar of national income not going to wages goes to capital: to owners, to shareholders, to the holders of the assets.

That number has been drifting downwards for forty years and the explanations for it are many, but the mechanism is not mysterious. When the cost of substituting for a worker falls, the worker's position in the negotiation weakens, and the share of the proceeds follows the position.

The asset that is not in the books

Here is the part that makes data the dream it is, and it is an accounting fact rather than an opinion.

Under the rules these companies report by, an intangible asset generated internally cannot generally be capitalised. A factory goes on the balance sheet. A fleet goes on the balance sheet. A dataset built from the behaviour of two billion people, which cost nothing to acquire, does not. It is expensed as it is created and then it simply is not there.

So the single most valuable thing these companies own does not appear among their assets. Their balance sheets understate them by design, which is one of the reasons the gap between book value and market value has become the widest in the history of markets.

Consider the properties of this asset, because they are unique. It costs nothing to acquire, because the users produce it as a by-product of being served. It does not depreciate with use; it improves, because more of it makes the predictions better. It cannot be stolen in any way that matters, because a copy without the system around it is inert. It has no residual value to anyone else and therefore no market price. And it compounds: the better the prediction, the more people use the service, the more data arrives.

There has never been an input like it. Capital has spent three centuries looking for an asset that does not wear out, does not need to be bought and gets better the more it is used. This is it.

Soon they will not need anyone at all

Consider also a change in the economy from knowing that soon conglomerates will not even need humans in order to be created, so it is feasible that every human being has their own, finally free to do what they please with their time. The economy will adapt to this. But it is the reality in plain sight of you and me and everyone: we are at the doors, and we are part of it, in the front row, and even creating existence.

In a short time, machines will be capable of having the brain structure in software that allows them to overcome any obstacle through the creativity of probabilistic data and the construction of scenarios. Come on, you know it is true.

The paths

That is not a statement of faith. It is a list of functions that used to require a department and a payroll, and which now require an account.

Designing the product and the brand. Writing the software. Writing the copy, in any language, at a quality that no small company could previously afford in more than one. Translating the entire operation into fifty markets. Answering customers at three in the morning in their own language. Keeping the books. Drafting the contracts. Researching a market before entering it. Building the campaign, placing it in the auction described in these pages and measuring what came back. Each of those was a salary, or several, or an agency. Each is now a line on a bill that is falling every quarter.

What is still hard is worth naming precisely, because it is where the remaining advantage sits. Anything that requires physical presence. Anything that requires a licence from a state. Anything that requires another human being to trust you, which still moves at human speed. And anything that requires capital up front, because a machine will design a product in an afternoon and will not pay for the first container of it.

So the paths are not mysterious and they are not equally open, but the list of things one person can now do without permission is longer than it has been at any point since the invention of the company.

Getting around the rules of advertising

The same capability applies directly to the one bottleneck named above, and this is where the creativity of probabilistic data and the construction of scenarios stops being abstract.

The conventional rule of advertising is simple and has held for a century: reach is bought, and whoever spends more is seen more. Creative work mattered at the margin; the budget decided the rest.

There are three places where that rule now bends.

The first is the price of a variant. Producing an advertisement used to cost money and days, so a campaign ran a handful of versions. When a variant costs almost nothing, a campaign can run thousands, each matched to a different segment and each measured. This matters because of how the auctions actually work: they do not simply sell to the highest bid. They weight the bid by how likely the audience is to respond, because a platform earns more from an advertisement people engage with than from one they ignore. Relevance therefore buys position, and relevance is now cheap to manufacture. That is the first time in the history of the business that creative output can be substituted, partially, for budget.

The second is the simulation. Before a single unit of currency is spent, a market can be modelled, an audience described, a message tested against a constructed scenario and the obvious failures discarded. The campaign that reaches the auction is the survivor of a tournament that cost nothing to run. Testing used to be the expensive part and it used to happen with real money in public.

The third is the part of distribution that was never sold. Algorithmic feeds distribute by predicted engagement rather than by spend, which means a sufficiently well-matched piece of content is carried for free by the same machinery that charges others for the same position. The difficulty was always that finding such a piece required many attempts, and attempts were expensive. They are not any more.

None of this abolishes the arithmetic at the end of this piece. The three largest intermediaries still take between sixty-two and sixty-four per cent of what the world spends on digital advertising, the auction still exists, and a very large budget still buys a very large amount of attention. What has changed is narrower and still significant: the price of competing for that attention without a very large budget has fallen, for the first time, in the same way the price of producing the thing being advertised has fallen.

In conclusion

In conclusion, socioeconomic levels will adjust somewhat, although immense capital will go on buying views that bring sales, and could surpass the rest.

That last clause deserves its figures, because it is the limit of the whole argument. The three largest intermediaries will take between sixty-two and sixty-four per cent of the roughly $740 billion spent on digital advertising this year. Attention is still bought, it is bought in an auction, and in an auction the deepest pocket does not always win but it never stops bidding. Cheap intelligence lowers the cost of making something excellent. It does not lower the cost of being seen.

Which is the asymmetry to watch over the next decade. Production has been democratised and distribution has not.

And then

Now all the power of a conglomerate is in a machine that can do everything without you doing anything. So what will you work on, out of pleasure or out of conviction, when you no longer have to work desperately in order to subsist?

That is not a rhetorical question and almost nobody has an answer ready. For roughly ten thousand years, the question of what a person does with their day was settled for them by the need to eat. Every structure we have, education, cities, the week, the idea of a career, retirement, was built around that settlement. If the cost of producing what a life requires keeps falling at the rate shown above, the settlement loosens, and the question arrives unprepared for.

The people who will find it easiest are the ones who already have an answer that has nothing to do with money, which is the quietest advantage in the whole of this arrangement and the only one that cannot be bought.

To conclude: intelligence and living cerebral freedom are the divine gift that makes us human. If we take advantage of it, it will be a better world for everyone. Until next time. And also: build hardware, guys!

Sources

Filed and published figures. Revenue per employee is calculated from the companies' own reported revenue and headcount in their annual filings. Labour's share of national income is the series published by the United States Bureau of Labor Statistics, which begins in 1947 and reached 53.8 per cent in the third quarter of 2025. The treatment of internally generated intangible assets follows the applicable accounting standards, under which such assets are expensed rather than capitalised.

Prices, identified as such in the text. Inference prices are published list prices of the providers and the market index compiled by Silicon Data; figures for the decline from roughly $30 to under $0.50 per million tokens are drawn from trade analyses of published pricing and vary with the model class chosen for comparison. Digital advertising totals and platform shares for 2026 are projections published by eMarketer and reported in the trade press.