A plain-language companion to the National Balance Sheet
Here is the thing almost nobody tells you about the federal budget: the government publishes what it owes every single quarter, and what it owns essentially never.
The debt number is famous. Everyone has seen the clock. The other side of the page — the roads, the ports, the labs, the water systems, the trained and healthy people who make all of it work — has no clock, no headline, and no line in the budget resolution.
So let's put one there.
The portfolio
America holds about $99.7 trillion in built capital. That is every road, bridge, factory, power line, house, server, and washing machine in the country, valued at what it would cost to replace them today, net of wear. About $21.3 trillion of that is publicly owned — mostly by states and cities, not Washington. The federal share is only $4.7 trillion.
Then there is the asset that dwarfs all of it. The earning power of the American people — what everyone alive today will earn over the rest of their working lives, discounted back to now — runs somewhere around $260 trillion. That figure is softer than the others and we'll come back to why. But the ordering is not in doubt. Roads and factories are the small part of what this country owns. People are the big part.
Then there's health — not hospitals, but health itself. When economists valued the longevity Americans gained after 1970, the answer came back at $95 trillion, and the annual flow of those gains ran around $3.2 trillion a year, roughly half of GDP, counted nowhere.
The National Park System is worth about $92 billion a year in what Americans say they'd pay to keep it — most of that from people who never visit and just want it to exist. Capitalise that and it's $3–5 trillion.
And then there's the entry that goes on the other side of the page, which we'll come back to.
Below all of that sit the things nobody has managed to price: the working landscape outside the parks, the institutions that make contracts worth signing, and the alliances that let America rent capability instead of buying it. All three are real assets. We're not going to invent numbers for them.
The problem
Assets wear out. That is not a policy position, it is physics. Concrete cracks, transformers fail, skills go stale, and every year some fraction of what the country owns has to be replaced just to stay level.
For sixty years, America has been replacing less than it uses up.
In 1968, federal investment ran at 6.2% of GDP. By 2018 it was 2.4%. Federally funded research was 1.86% of GDP in 1964 and 0.70% by 2020 — and the federal share of all American research fell from two-thirds to about a fifth.
Over the same stretch, the debt went up. It's on course to run from about 101% of GDP this year to 175% by 2056.
Put those two facts next to each other and you get the actual problem, which is not the one people usually argue about. Borrowing money to buy an asset is a trade: the debt goes up and so does what you own. Borrowing money to cover this year's expenses is something else. You end up with the debt and nothing to show for it.
America has spent sixty years doing more of the second thing and less of the first. The engineers' assessment says the same thing in their own language: bringing American infrastructure back to good working order needs $9.1 trillion over ten years, about $5.4 trillion is coming, and the gap is $3.7 trillion.
What this stuff actually returns
Numbers as large as these stop meaning anything, so here are four programs with real price tags.
The Interstate Highway System cost $128.9 billion by the final government estimate in 1991. One study that modelled the American economy with and without it put the difference at $742 billion a year. Whatever you think of the precision, the order of magnitude is not close: the system returns something like its entire construction cost every couple of months.
The internet was seeded by the federal government for $124.5 million — ARPANET, CSNET, the NSFNET backbone, and the programs that connected universities to it. That is the narrow figure, covering the networks themselves rather than the decades of computing research behind them. It remains, on any accounting, the best money the United States has ever spent.
The electric grid would cost roughly $5 trillion to replace. Seventy percent of its power transformers are twenty-five years or older. New ones now take between eighteen months and four years to get, up from about a year in 2021.
Freight rail got none of this. American railroads are privately owned and have put about $840 billion of their own money into the network since 1980 — close to $1.4 trillion in today's dollars, with no taxpayer capital. They move around 40% of the country's long-distance freight at rates 44% cheaper, after inflation, than in 1981.
That last one matters for a reason we should be straight about.
The train argument, honestly
You have seen the comparison. China's high-speed rail network passed 50,000 kilometres at the end of 2025 and carried 4.28 billion passenger trips that year. The United States runs nothing at those speeds. The photograph is real and the gap is real.
The photograph is also half the ledger. America did not fail at rail. America chose private freight over public passengers, and on freight that choice has been vindicated about as completely as a policy choice can be. Europe moves people by train and goods by truck. America does the reverse, and on fuel and emissions the American arrangement comes out ahead.
The real indictment is narrower and worse. It is not that America chose roads. It is that having chosen roads, America stopped paying to keep them. Building one network superbly and then declining to maintain it is not a strategy. It is an accounting error that has been running for two generations.
So is the investment worth it?
We ran fifty years of arithmetic on it — 2026 to 2076, three different renewal rates, every assumption written down. Here is the answer, including the part that is less flattering than we'd like.
Public infrastructure returns about two dollars for every dollar. Not ten. Two, maybe a bit more, and under pessimistic assumptions closer to one and a half. Two-to-one over fifty years is a good investment and it comfortably beats what the money would otherwise do. It is not a miracle, and any page promising you tenfold returns on pipes and pavement is selling something.
On the current path, the public capital stock shrinks against the economy it has to carry — from 0.70 times GDP today to 0.59 by 2076. It doesn't shrink in dollars. It shrinks in capability, quietly, one deferred maintenance cycle at a time. That is what liquidation looks like when you draw it instead of describing it.
The tenfold returns are real. They're just somewhere else. A dollar of research generates more than ten dollars of benefit across the economy. The Perry Preschool follow-up found an annual rate of return of 7–10% — an interest rate, not a multiple — which works out to seven to twelve dollars back per dollar once you add up a lifetime of benefits and discount them.
That last distinction matters more than it sounds. Perry was a small program for very disadvantaged children in 1960s Michigan, and its return is an average on the group with the most to gain. A national program spends its last dollar on a child who'd have been fine anyway, and that dollar returns much less. The Accord books child investment at 1.1–1.2x on the marginal dollar, not at Perry's headline. Being disciplined about that gap is what earns the right to cite Heckman at all.
Which brings us to the finding that should bother everyone.
The bill nobody books
Here's the entry that goes on the other side of the page.
Carbon pollution has a federal price. The EPA's current methodology puts the damage from one ton of CO₂ at $190, with a range of $120 to $340 depending on how much weight you give to future harm. America emitted 4.9 billion tons of energy-related CO₂ in 2025.
Multiply, and discount over the same fifty years as everything else: $29.3 trillion.
That is 1.4 times the entire public capital stock of the United States. It is more than the total fifty-year cost of the investment program described above. The country is carrying an off-book liability larger than the on-book asset — and larger than the bill to fix the on-book asset.
Two honest caveats. This is the most politically contested number in federal economics; the working group that produced it was disbanded in 2025 and the EPA was told to reconsider. And even at the low end of the range you get $18.5 trillion, which changes nothing about the conclusion. It also counts carbon dioxide only — methane and nitrous oxide would push it higher.
The uncomfortable part
Line up America's assets by how well they're measured, then line them up by how much they return. The two lists run in almost exactly opposite order.
Roads and bridges are measured beautifully and return about two to one. Research returns more than ten and gets counted as an expense. Children return more than anything else on the list and appear nowhere. Institutions and alliances return an amount nobody has ever seriously tried to calculate, which is precisely why they're the first things cut and the hardest to defend in a hearing room.
A country that manages only what it measures will, over sixty years, systematically starve its highest-return assets. That is not a hypothesis about what might happen. It is a description of what did.
What we are not claiming
The $260 trillion human capital figure comes from applying a method used across advanced economies rather than from a purpose-built American calculation. Treat it as an order of magnitude.
The fifty-year scenarios model roads and pipes and wires only. We did not take the research and early-childhood returns and project them across the whole country, because scaling results from a small trial to a national program is not something the evidence supports. Those numbers stay where they belong: as evidence about where the high yields are, not as a forecast of what a national program would deliver.
The health capital figure stops at 2000 and isn't inflation-adjusted. Worth saying plainly: American life expectancy has stalled and then fallen since about 2010, so while the accumulated stock stands, the annual flow probably doesn't — and may now be running negative. Health is the one line where this country may currently be posting a loss.
The parks number comes from asking people what they'd pay, and people reliably say more than they do. It also covers 85 million of roughly 640 million federal acres, so it's a floor on a floor.
And the assets with no price stay blank. Making up a number for America's working landscape, its institutions, or its alliances would be worse than admitting we don't have one. The gap is real, and the honest thing to say about it is that every one of those errors runs in the same direction: this analysis understates what the country owns and what it stands to lose.
The bottom line
America is not deciding whether to build a hundred-trillion-dollar portfolio. It already owns one, built mostly by people who are dead now, and it is currently spending it down.
The choice on the table is the renewal rate. At about two to one on the concrete and considerably better on the people, the arithmetic supports raising it. And the rate the New American Accord proposes — 4.4% of GDP in public investment — is still below where the country ran in 1968, in the years it was building the thing we are now living off.
The full analysis, with charts, tables, scenario specifications and sources, is on the National Balance Sheet page. The China comparison gets its own treatment there and a fuller one shortly.