Capital Has a Shape
The power law built the software era. It cannot build the physical one, and nobody has invented the instrument that can.
The transcontinental railroad was built on government land grants and federal bonds. The interstate was a federally funded defense program. The integrated circuit was bought into existence by two customers, the Apollo program and the Minuteman missile, which together absorbed nearly the entire early output of the industry and drove the price of a chip down by orders of magnitude. The internet was a DARPA project. For roughly a century the West financed its hardest physical bets with capital that could wait decades and absorb the risk that the thing might not work at all.
Then it stopped. The standard explanation is that we lost our nerve, drowned in permitting and litigation, stopped believing in the future. That explanation is everywhere, and it mistakes a symptom for the disease. The West did not stop building atoms because of culture or regulation. It stopped because the dominant instrument for funding ambition changed, and the new form cannot hold atoms.
Thiel’s long-beaten line was that we wanted flying cars and got 140 characters. The machine that allocated the West’s risk capital for forty years delivered exactly what its return structure rewards, and what it rewards is bits. Capital has a shape, and its shape selects which futures get funded.
More Money, Fewer Bets
Venture capital runs on a power law. A fund makes many bets, expects most to fail, and needs a few to return the entire fund many times over. That math has one ruthless implication. A venture-backable company has to get enormous, fast, on relatively little capital. It has to scale without building much, and approach a monopoly before the fund’s clock runs out. One kind of business reliably fits: software. The power law is a filter that admits software, and the rare physical company that can mimic software's economics, and rejects the rest, because almost nothing physical gets that big that fast that cheap.
The filter is not failing. It is working harder than ever. In the first quarter of 2026, global venture funding hit roughly $300 billion, an all-time record. American companies took about 83 percent of it. Around 80 percent went to artificial intelligence. Five deals accounted for nearly three-quarters of the entire US total. This is the power law in its purest form: the West’s risk capital, more abundant than at any point in history, concentrating into a handful of names, all building bits.
Look closely at the largest of those rounds and they are not venture bets at all. They are sovereign wealth funds, strategic corporates, and a few crossover funds writing ten-figure checks into the safest trade in the market. The instrument is already straining past its own structure, reaching for capital it was never built to use. It is doing all of that for bits. When the frontier turned physical, the same instrument flinched.
Where Venture Flinches
The reason is not that atoms lack a power law. Atoms have ferocious power laws. The first company to crack a manufacturing process at scale takes the category and holds it. The problem is the clock. The atoms power law often plays out over fifteen years or more, and a venture fund has ten. The fund has to return capital on a schedule the physical world does not honor, so it cannot wait for the curve that would pay it.
Beneath the clock is a deeper problem in how the risk is priced. A first-of-a-kind physical project stacks two unrelated risks into one number. The first is technology-execution risk: will this novel process actually work at commercial scale, the first time, in steel and concrete. The second is ordinary asset risk: assuming it works, will the plant run profitably for thirty years. Fused into a single number, that risk is too uncertain for venture equity and too unproven for project-finance debt. Each side looks at the blob and walks. The result is the most dependable bottleneck in the physical economy. In a 2025 survey of climate-tech investors, a majority named first commercial-scale facilities the single hardest stage to finance, and most expected that capital to shrink further. The problem, as one of them put it, is not a shortage of money. It is how the money is organized.
Concede the part that is real. Design and simulation are commoditizing fast under AI. Execution and scale-up are not, and never will be, because building a thing for the first time is irreducibly hard. That irreducibility is the entire bottleneck.
This is not nostalgia for a kind of capital that cannot return. The patient model still builds atoms today. In December 2025, China launched a national venture fund with a twenty-year life against the standard ten, and called it patient capital, explicitly for development cycles private money will not sit through. The instrument exists. It is simply not ours.
And ours is not quietly fixing itself. Venture debt, continuation vehicles, the hybrids of the last few years, all reshuffle the capital stack. None of them extends the clock or shrinks the irreducible tail. They are cosmetic by construction.
Hold the Risks Apart
The answer is not more patience or more money. It is a different structure. Stop pricing the two risks as one thing; tranche them. Put risk capital on the technology layer, the bet on whether the physics works. Put project-finance debt on the asset layer, the bet on whether a proven plant runs. The instrument’s whole job is to hold the two risks apart, so each is priced by the capital built to price it, instead of fused into a number that repels both.
One tranche resists this. The first-of-a-kind tail carries a risk no private investor will hold alone: the catastrophic, binary chance that the process fails outright, or that even working, it meets no market able to repay it. Neither can be diversified or modeled away. The state cannot make the physics work, and it has no business guessing which physics will. What it can do is sit behind private capital that has already chosen the bet and staked its own money on it, and take that catastrophic slice off the table. That is what makes a private investor willing to underwrite an unproven process at all, and the order is the discipline: the state never funds what private capital has not already selected and backed, so the picking is never the government’s to do.
The cleanest form gates itself. An offtake, a commitment to buy the first plant’s output, pays nothing unless the plant produces, so it cannot rescue a project that fails. A loan guarantee or a first-loss tranche needs the discipline bolted on, released against milestones so the money follows demonstrated progress, not promises. Each lowers the return a genuine bet must clear. None rescues a bet no private allocator wanted. A bad idea dies exactly as it does now, unfunded. The state takes only the loss markets refuse to underwrite, gated behind private conviction and paid against results, and picks nothing.
Solar already proved the easier half of this. The technology was mature for years before it scaled. What unlocked it was not a breakthrough in physics but a sequence of financing structures matched to its shape. The federal investment tax credit de-risked the first dollar. Tax-equity partnerships moved that credit to investors who could actually use it. The power purchase agreement turned a panel into a thirty-year contract a bank would lend against. Securitization bundled those contracts into something the capital markets would buy. Every one of those structures existed to put one more gigawatt of panels onto roofs and into fields. The financing was the innovation. The technology was almost boring.
But the harder half, funding the tail before anyone knows it works, has its own proof, and this essay opened with it. When Apollo and Minuteman committed to buy nearly the entire early output of the integrated-circuit industry, that was not simple procurement. It was a state offtake that removed the one risk no private buyer would absorb: that even a working chip might find no market large enough to repay the fab. The technical risk stayed with the firms. The state did not design the circuit or pick the winners. It paid only for chips that worked, guaranteed the demand if they did, and let private capital bear the rest and race down the cost curve.
The opposite case is just as sharp. Pure state allocation, the kind that picks the projects too, is the wrong instrument, and China is the proof as much as the warning. Its state-led funds have returned less than a third of what private equity does. Companies it takes public run about two percent lower on return on assets for two years after listing. Its flagship national chip fund froze for five months after a corruption scandal took down its leadership. The same vast state capital that builds the plant is the capital that buries the balance sheet, because a structure that funds without discipline is a structure that rots. The most careful read of the data lands in the same place: companies backed by partially government-owned managers reach the public markets at meaningfully higher rates than those backed by wholly government-owned ones. The state at the edges works. The state at the center does not.
Right Impulse, Wrong Instrument
The West is reaching for state capital in real time, which kills the objection that any of this is politically unthinkable. In February 2025 the President signed an executive order to build a US sovereign wealth fund. In April 2026 OpenAI proposed a public wealth fund to give every citizen a stake in AI. Now, in June 2026 Bernie Sanders went further, proposing a one-time fifty percent tax, paid in stock, on OpenAI, Anthropic, and xAI, to seed exactly such a fund, with government board seats and voting shares attached. A socialist senator and a Republican president, plus the labs themselves, all want the same thing. The will to use state capital has arrived.
It is aimed at exactly the wrong target. Every one of these proposals is about owning the bits frontier and handing it around: taking equity in AI companies that have no trouble raising hundreds of billions on their own. It is the wholly-state instinct in its crudest form, pointed at the one frontier that least needs help, and it does nothing to tranche the risk of a single physical plant. The West has finally found the nerve to deploy public capital, and it is about to spend it on ownership instead of creation.
So the real question for the next twenty years is not whether the West uses state capital. It will. The question is whether it builds the instrument or builds the Bernie fund.
Who Holds Power
Build the instrument, and you get the one thing the wholly-state model can never deliver: the patience of sovereign capital without its waste, because the state underwrites only the risk markets refuse while private capital still makes every call. That is the entire trick. Buy China’s time horizon without buying China’s malinvestment.
Power will not go to whoever merely designs the structure. Financing innovations have no famous authors. No one can name who invented the power purchase agreement or the leveraged buyout, and everyone can name the firms that got rich deploying them. Power goes to whoever controls and deploys the instrument at scale. The operator who designs it and runs the capital through it takes the category, the way a monopoly is meant to be taken, by owning a structure no one else can copy.
That operator will not come from inside today’s money. The people closest to the spigot, the megafund LPs and the franchises built on the power law, are structurally the wrong builders, because the entire apparatus that makes them rich is the fund-clock model the instrument has to escape. They cannot build the thing that obsoletes their own return math. The seat sits outside the current perimeter, which is exactly why it is open. It is the one position the incumbents cannot take, because taking it contradicts their own incentives. The winning archetype is a balance sheet with sovereign-scale patience and a tolerance for the unhedgeable tail, fused to private-market allocation discipline. Neither a fund nor a state. It has no clean name yet, and it has no name precisely because the only people positioned to name it are the people who cannot build it.
Watch where the bottleneck bites hardest to see what the instrument unlocks first. Power is the binding input to everything physical now, and it is getting scarcer, slower, and more expensive, not cheaper. The cost of a new gas plant has risen by roughly two-thirds since 2023, and turbine waitlists now stretch into the next decade. Point-of-use programmable power, generated where it is consumed, is the paradigm case for the new instrument, because it is the perfect blob: a novel-technology tranche sitting on a deployed-asset tranche, exactly the risk current capital cannot price and the instrument is built to.
And it is only the first. The instrument is generative: the industries we cannot yet name are downstream of the instrument we have not yet built.
The counterfactual is the cost of getting this wrong. If the West spends its one moment of willingness on the Bernie fund, the bottleneck holds, the physical frontier gets built elsewhere or not at all, and a generation that could have owned the next economy ends up renting it.