Who Owns the Machine? ⚙️

Not what the machine can do — whose name is on it. This is not a political verdict; it’s a walk around the ownership question of the century, with the strongest voices from both sides given the floor.

---

The Server Hall and the Kitchen Table 🏭

Drive an hour past the edge of several American cities this summer and you’ll find the same scene: a building the size of a shopping mall going up in a field, windowless, humming before it’s finished, ringed by substations. Inside, eventually: racks of processors that may end up doing a meaningful share of the region’s thinking — drafting its contracts, reading its scans, tutoring its kids.

Now hold that building in one hand, and in the other hold a kitchen table a few miles away — the one where a family is doing envelope math, the way the 37% of American adults who told the Federal Reserve they wouldn’t cover a $400 emergency with cash or its equivalent do math.

Here is the question this essay exists for — not a left question or a right question: when that building starts producing a large share of the economy’s value, does anyone at that kitchen table hold a claim on it — or just a subscription to it?

A deed versus a login. The entire scarcity age turned on that distinction — and the last time a machine rewrote the economy, the era’s wisest observers converged on it too. First, the facts.

---

The Ownership Question of the Century 🗺️

Why now? Because the ownership structure of the new productive base is being poured like concrete — fast.

The numbers, plainly. Training a single frontier AI model now costs hundreds of millions of dollars, with costs growing roughly 2.4× per year, per Epoch AI — on that trend, billion-dollar training runs arrive within a year or two. The four largest technology companies plan roughly $725 billion in AI infrastructure spending in 2026 alone. And the computing substrate underneath nearly everything already runs through a needle’s eye: three companies hold nearly two-thirds of the world’s cloud infrastructure market.

If the optimists are even half right — if machine intelligence becomes the productive base of the economy the way land and factories once were — then “who owns compute, models, energy, and data” is not a tech-policy question. It is the property question of the century, being answered by default.

And yet — honesty requires the counter-fact in the same breath — the picture is not simply a closing fist. Open-weight models, whose parameters anyone can download, run, and modify, have genuinely proliferated: Meta’s Llama line, Mistral, Alibaba’s Qwen, DeepSeek. Capability that cost nine figures to create can now, in meaningful if lagging form, run on hardware a small business owns outright. Concentration and diffusion are both happening, at once, at speed.

One more thing — evidence for the series’ premise: this issue scrambles the jerseys. Open-source AI is championed by libertarians and left-technologists. Skepticism of Big Tech runs from Josh Hawley to Elizabeth Warren — senators who agree on almost nothing else. A question that refuses to sit on either bench is usually older than the teams. This one is.

(Full disclosure: this publication is itself produced with AI tools — a reason for humility here, not authority.)

Now the steelmen — real ones.

---

The Strongest Case from the Left 🧣

The left’s best argument is a serious one.

First: concentration of a general-purpose technology is different in kind. When one firm dominates sneakers, you buy other sneakers. When a handful of firms own the technology every other business must route through, their position starts to resemble infrastructure — and infrastructure has always drawn public claims. That’s the American antitrust tradition itself: the Supreme Court broke up Standard Oil in 1911 under the Sherman Act; the government unwound AT&T’s telephone monopoly in 1984 — both wrenching, both now broadly credited, including by many market conservatives, with releasing waves of competition. Refusing private chokepoints on the economy’s arteries is not radical. It’s older than the radio.

Second: public investment deserves acknowledgment upstream of private returns. The field was carried for decades by public money — DARPA funded and legitimized AI research from the 1960s onward and built the ARPANET that became the internet; NSF, NIH, and NASA carried supporting roles through the unfashionable decades. The taxpayer took the earliest, least fundable risk. When the payoff lands concentrated in a few private balance sheets, “the public should hold some claim on what the public seeded” is not envy. It’s accounting.

Third: if capital starts doing the cognition, labor’s share of income — already falling for decades — has no obvious floor. The wage was how ordinary people held a claim on the economy’s output. If machines absorb the tasks, the claim thins precisely when the output explodes. A democracy whose productive base answers to a handful of boards, while most citizens hold no stake, has a legitimacy problem no earnings call can fix.

That’s the left’s case, honestly: chokepoints, public seed capital, and the evaporation of labor’s claim. It deserves a real answer, not an eye-roll.

---

The Strongest Case from the Right 🧢

Now the other steelman — also strong.

First: this concentration is what building the thing actually looks like. Somebody had to risk staggering sums on an unproven technology — and it wasn’t a ministry. It was private firms, spending hundreds of billions of shareholders’ dollars a year with no guaranteed return. The scale critics call a moat is, from inside the engine room, the price of admission physics is charging. If your response to “private actors funded the most expensive R&D project in history” is to punish the funders, expect less funding of the next one.

Second: the government’s record at picking winners is poor, and antitrust adventurism has real casualties. Breakups look clean in retrospect and chaotic in the moment; a decade of litigation aimed at firms mid-race can kneecap the racers without helping a single household. And there’s a national-security floor here that serious people across parties accept: the frontier is contested internationally, and a country that hobbles its own builders may simply cede it to rivals with no such scruples.

Third — the right’s sharpest, least appreciated point — regulation is the incumbent’s best friend. Big firms can afford compliance departments; garage challengers cannot. This isn’t cynicism; it’s the capture theory George Stigler formalized in 1971 — regulation tends to be acquired by the regulated and operated for their benefit. The pattern is already knocking in AI: when OpenAI’s chief executive asked Congress to create a licensing agency for powerful models, senators pressed him on exactly this danger — rules only the wealthy can meet — and he conceded it was real. Rules written to tame giants have a long record of entrenching them. If you fear five companies owning the future, be very suspicious of policies whose practical effect is that no sixth can ever afford to exist. And note the alternative’s pedigree: open markets plus secure property rights built every previous round of abundance. The burden of proof sits on those proposing to rewire that.

That’s the right’s case, honestly: the engine is fragile, the state’s hands are clumsy, and the cure has historically fed the disease.

---

The ReThink: Widen the Claim 🌱

Hold both steelmen at once and notice: they’re arguing about control, and mostly talking past claims. “Nationalize it” and “leave it alone” are both answers to who runs the machine. The older, better question is whose name is on its output.

This series keeps one measuring stick; grade openly. Does a proposal (a) move toward abundance, (b) put the most vulnerable at the center, and (c) keep the both-and — market engine plus dignity floor? Fail any leg, fail the test — whatever the jersey.

Here we have the rare luxury of precedent. In 1891, at the bottom of the industrial upheaval, Leo XIII’s Rerum Novarum rejected both collectivizing the factories and blessing their concentration, and landed on a third thing: “The law, therefore, should favor ownership, and its policy should be to induce as many as possible of the people to become owners.” Don’t seize the machine. Don’t worship it. Widen the claim on it. Our essay on newborn investment accounts picked up this thread; this essay is where it was always headed.

What does “widen the claim” look like in 2026? Proposals to debate — not a platform:

1. A citizen’s stake in the windfall — the Alaska pattern. Alaska, one of the reddest states, voted in 1976 to bank its oil wealth for every Alaskan: royalties feed a sovereign wealth fund, which since 1982 has paid every resident an annual dividend — $1,000 in 2025, $1,702 the year before. Note what it is not: not state ownership of the oil companies, not means-tested welfare, and no dent in Alaska’s market economy. A universal claim on a windfall, riding on top of private production. If AI produces the windfall its own builders forecast, a sovereign-wealth-style public stake — funded by equity set-asides or windfall-contingent commitments rather than engine-choking taxes — is the same species of idea. Researchers at the Centre for the Governance of AI sketched a voluntary version, the “Windfall Clause”: firms pre-commit a share of extreme, transformation-scale profits to broad benefit. Grade it: abundance-compatible, vulnerable-centered (universal, no forms to fumble), both-and by construction. Debate the mechanism hard; the shape passes.

2. Open weights as ownership diffusion by other means. The reframe hiding in plain sight: an open-weight model is redistributed capital. When a frontier-adjacent model can be downloaded and run by anyone, the capability the nine figures bought stops being excludable. A clinic in Nairobi, a plumber in Ohio, a school district that owns its own tutor and its own data: not charity access — holding the means of cognition. This is why the jerseys scramble here: it’s simultaneously the most libertarian answer (no state required) and the most radical one (the capital is simply… shared). The implication either camp could carry: don’t regulate open weights out of existence to soothe incumbent-shaped fears. Grade it: abundance ✓, both-and ✓; vulnerable-centered only if paired with access and know-how — weights on a server help no one who can’t run them.

3. Compute as the land-grant question. In 1862, the Morrill Act took the era’s productive base — land — and granted 30,000 acres per congressman to seed public colleges, so farmers’ and mechanics’ children could hold the era’s capability, not just supply its labor. It didn’t nationalize agriculture; it widened the on-ramp. The 2026 rhyme: public compute reserves for researchers, students, and small firms — already live in miniature as the NSF’s NAIRR pilot, which the CREATE AI Act — sponsored by two Republicans and two Democrats in the Senate — would make permanent — so the frontier stays contestable and capability doesn’t pool solely where the capex is. Grade it: keeps the private engine, widens who can challenge it, gives the kid at the far end of the table a door that doesn’t bill monthly.

The common thread: each keeps the engine — private frontier competition, property rights, prices — and writes the floor: a universal claim (a dividend, a download, an on-ramp) rather than a permanent subscription. The one-sentence test for every future proposal here: does the child at the far end of the table end up holding a claim on the machine, or just a login that can be revoked?

---

Honest Tradeoffs, Both Directions ⚖️

An un-stress-tested proposal is a brochure. Both directions:

Public stakes can rot into politics. A sovereign fund is only as good as its insulation; a dividend fund raided for budget holes becomes exactly the clumsy-hands problem the right warned about. Alaska’s own dividend is now an annual political fight. Design matters more than intention.

Open weights cut both ways. The same non-excludability that makes open models “redistributed capital” also hands capability to people building scams, weapons know-how, or surveillance tools. Serious open-weight advocacy carries the misuse ledger honestly, and there may be capability thresholds where diffusion’s math genuinely changes.

Dividends draw the dependency critique — from both benches. The right worries a check erodes the dignity of earned work; parts of the left worry it becomes hush money that replaces structural claims — good schools, real bargaining power — rather than supplementing them. Both worries have teeth. A claim on the machine should fund agency, not purchase quiescence.

And the timing is unknowable. Every mechanism above is sized to a windfall that may arrive slower, stranger, or smaller. Building distribution machinery for profits that never materialize has a cost — and so does waiting until the concrete has fully set. Nobody gets to be certain here, including us.

---

What to Carry Out of the Room 🧭

If you shape rooms — a company, a portfolio, a committee, a classroom — three things worth carrying:

One: refuse the false binary early. The debate will be sold to you as nationalize-versus-laissez-faire. That’s the thermostat fight from a previous century. The live question is claims, not control.

Two: watch the ratio of claims to subscriptions in whatever you build or fund. Every architecture decision — open or closed, owned or rented — is a small vote on whether the abundance has other people’s names on it. The 1891 insight was never just for legislatures.

Three: keep both nightmares on the table, because both are real. Abundance controlled by five companies is just scarcity with better graphics — and an engine smothered by its referees builds no abundance at all. Seeing only one of those failure modes is half-blindness, whichever half you vote with.

The machine is being built either way. The question with your name on it is quieter: when it hums, who does it hum for?

---

Honest Fine Print 📎

This is not a political verdict. Serious people on the left fear the chokepoint; serious people on the right fear the clumsy hand; both fears have historical receipts. We’ve tried to give each side its strongest voice — and if you finish unable to tell which way this essay votes, it worked.

This is not investment advice. Nothing here recommends buying, selling, or holding anything.

The trajectories are deeply uncertain. Forecasts about this technology have a rich graveyard. Cost curves could bend; open models could stall or leap; the windfall could be a trickle. We’ve cited what’s checkable and held the rest loosely.

Proposal, not decree. The mechanisms above are offered the way you’d slide a napkin across a table — to be argued with, redrawn, and improved by people who disagree with each other. That’s the point of a long table.

---

This essay closes the first ReThink Policy set. Facts current as of July 2026.

— The ReThink · firstfruits 🌱 · truth first, hope on top