This Saturday, the fireworks will be bigger than usual. America turns 250 years old on July 4, and the country will mark its semiquincentennial the way it has marked every birthday since the first one: Flags on porches, burgers on grills, kids with sparklers, families sprawled on blankets waiting for the sky to light up. For all of its imperfections, and it has plenty, this country remains one of history’s great experiments in individual freedom and entrepreneurial ambition. That is worth celebrating, and I intend to celebrate it.
But somewhere between the parade and the barbecue, it is worth pausing on a quieter story. The most important thing about America’s 250th birthday is not what happened in 1776. It is what the country is building in 2026.
Drive through central Ohio, northern Virginia, west Texas or the high plains of Montana and you will see it: Cranes on the horizon, substations rising out of farmland, concrete pours that run around the clock, transmission corridors being surveyed, fiber trenches being cut, and cooling plants taking shape next to buildings the size of aircraft carriers. It does not look like a revolution. It looks like a construction site. That is usually how the big ones start.
Every Generation Builds Something
Every generation of Americans has had a defining infrastructure project. The canals opened the interior of a young country and turned the Erie corridor into an economic artery. The railroads stitched a continent together and created the modern industrial economy almost as a byproduct. Electrification transformed factories, farms and homes over the course of half a century. The interstate highway system reorganized where Americans lived, worked and shopped. Telecommunications, the internet and then cloud computing each did the same thing for information that highways did for goods: They collapsed distance and made scale available to anyone with ambition.
Today’s project is different in kind but identical in ambition. Across the country, hundreds of billions of dollars, and soon trillions, are flowing into AI factories, semiconductor fabrication plants, power generation, transmission lines, networking, fiber, cooling systems and the software platforms that tie it all together. The five largest hyperscalers alone are on track to spend roughly $650 billion or more on capital expenditure in 2026, with analysts projecting the figure surpasses $1.1 trillion in 2027. Goldman Sachs now estimates the total build-out across compute, data centers and power could consume about $7.6 trillion between 2026 and 2031. Our colleagues at Futurum peg the 2026 hyperscaler number at $660 to $690 billion, nearly double 2025 levels, with roughly three-quarters of it going directly into AI-specific infrastructure.
Those numbers are so large they stop meaning anything, so let me put them another way. Amazon’s capital budget for this one year is the largest single-company technology capital commitment in history. The combined hyperscaler spend for 2026 exceeds the annual GDP of most countries on Earth. And that is just the private cloud players. It does not count the chip fabs going up in Arizona, Ohio, Texas and New York, the utility-scale power projects, the restarted nuclear plants, or the small modular reactors on drawing boards from coast to coast. America is not building data centers. America is building an Intelligence Grid.
The Electrification Parallel
The analogy I keep coming back to is electrification, and I think it holds up better the harder you push on it.
When the first central power stations went up in the 1880s and 1890s, electricity was a luxury product for factories and wealthy urban districts. Nobody fully understood what a nation wired for power would become. Over the following decades, businesses electrified their operations one motor and one assembly line at a time, and in doing so they laid the foundation for the entire twentieth-century economy without ever framing it that way. Electricity stopped being a technology and became an assumption. You flip the switch, the lights come on, and you never think about the generating station a hundred miles away.
Something similar has already happened once in our own industry. For thirty years, enterprises digitized their operations, moved workloads to the cloud, and instrumented everything with data, mostly for reasons that had nothing to do with artificial intelligence. Nobody in 2005 was building a data lake because they anticipated large language models. But all of that digitization turned out to be the substrate AI needed. We laid the foundation without knowing what we were laying it for.
Now the pattern is repeating one level up. The intelligence infrastructure being built today will, if this build-out succeeds, eventually become as invisible and as essential as the electrical grid. Intelligence becomes a utility. It is simply there, metered and reliable, available whenever and wherever it is needed, and nobody thinks about the GPU cluster behind the API call any more than they think about the turbine behind the light switch.
What Makes This One Different
Skeptics will point out, fairly, that we have heard versions of this speech before. Every technology cycle gets compared to electricity eventually. But there is something genuinely distinct about this one, and it is worth being precise about what it is.
Previous technology waves consumed primarily capital and talent. This one is consuming capital, land, electricity, water, skilled trades, grid capacity and public attention simultaneously, and at levels that are straining every one of those systems at once. Deloitte estimates AI data centers will need 92 gigawatts of additional electric power by 2027, power that in many regions simply does not exist yet. Grid interconnection queues in the United States have ballooned past 2,100 gigawatts, more than the total capacity of the existing grid, and industry analysts project 30 to 50 percent of planned 2026 data center capacity will slip to 2028 because the power and the transformers are not there. One analysis found the U.S. would need to purchase 90 percent of global semiconductor output over the next five years to support all of the announced data center load.
That is not a technology trend. That is an organizing principle. Governments are rewriting permitting law around it. Utilities are rebuilding decades of demand forecasting around it. Investors are reallocating trillions around it. Entire regional economies, from Abilene to Central Ohio to rural Montana, are being reshaped by it. No previous wave of enterprise computing, not client-server, not the web, not even cloud, ever bent the physical world this hard.
Who Actually Captures the Value
History offers a caution here, and it is one that the companies writing those trillion-dollar checks should sit with.
During electrification, the utilities built the grid, but the great fortunes of the electrical age were mostly made by companies that built businesses on top of cheap, ubiquitous power: Appliance makers, aluminum smelters, broadcasters, manufacturers who reorganized entire factories around the electric motor. The utilities themselves became regulated, essential and only modestly profitable. During the internet era, the telecom companies laid the fiber, spent staggering sums doing it, and watched Amazon, Google and a generation of companies that did not exist yet capture most of the value created on top of their networks. Some of those telecoms never recovered from the build-out.
Could the same pattern emerge again? It is at least possible that the companies spending these vast sums on AI infrastructure end up resembling regulated utilities, providers of affordable intelligence as a service, essential and commoditized at the same time. The truly transformative wealth may accrue instead to companies that invent entirely new industries on top of abundant intelligence, industries we cannot name yet because they do not exist yet. Nobody standing in a power station in 1900 predicted commercial aviation, broadcast media or cloud computing. Whoever is standing in an AI factory in 2026 is probably just as blind to what gets built on top of it by 2076.
I do not say that to diminish the builders. Somebody had to string the wires. But the lesson of American infrastructure is that the grid and the fortune are rarely the same thing.
The Commercialization Advantage
There is a deeper historical theme here that deserves more attention than it gets, especially this week.
America has rarely won because it invented something first. The transistor, the internet protocols and even much of the foundational AI research came out of labs, universities and government programs, and plenty of other countries have produced brilliant science. What America has done better than anyone, for going on 250 years, is commercialize innovation at scale. The legal framework, the depth of capital markets, the entrepreneurial culture, the bankruptcy laws that let founders fail and try again, the willingness to let disruption actually disrupt: Those are the things that repeatedly turned laboratory breakthroughs into global industries. The railroad was not an American invention. The American railroad economy was.
The honest question as the country turns 250 is whether those advantages are still intact. Capital is certainly still here, in historic quantities. The talent still comes when we let it. But the environment for experimentation is more contested than it has been in a long time. Government is more deeply involved in technology than at any point in my career, through export controls, industrial policy, procurement, safety regulation and direct investment. Some of that involvement is warranted, and I have argued as much in these pages. National security concerns around advanced compute are real. But there is an unavoidable tension between a state that steers technology and an economy that discovers it, and both parties, in different ways, are pulling government deeper into the steering role. The challenge for the next decade is holding two ideas at once: technological leadership matters, and so does preserving the permissionless experimentation that produced it in the first place. Countries that manage only one of those tend to regret it.
The Concentration Question
And then there is the question of power, in the other sense of the word.
The intelligence economy is being built by a remarkably small number of companies. A handful of firms control the frontier models, the cloud infrastructure, the leading-edge semiconductor ecosystem and an extraordinary share of the capital being deployed. The same names appear on both sides of most of the major transactions, investing in each other, buying from each other, hosting each other. We have seen concentrations of economic power like this before in American history, and the comparison to the Gilded Age gets made so often it has become a cliché. Clichés usually contain something true.
The railroad barons and the oil trusts also built genuinely essential infrastructure, and the country was genuinely better for having it. The problem was never the building. The problem was what concentrated control over essential infrastructure eventually did to competition, to pricing, to political influence and to the ability of the next entrepreneur to get a fair shot. It took decades and a generation of trust-busters to work out a settlement between infrastructure at scale and markets that stayed open.
I am not going to pretend to know whether the intelligence economy will require a modern version of that settlement, or what new antitrust thinking for AI would even look like. Existing doctrine, built around consumer prices, maps awkwardly onto services that are often free and inputs that are compute, data and talent. But the question deserves to be on the table now, while the concrete is still being poured, not after the market structure has hardened. If intelligence really does become the essential utility of the next American century, then who controls it, on what terms, and at what price will be one of the defining political economy questions this country faces. Pretending otherwise would be naive, and naive is one thing 250 years of history should have cured us of.
The People Doing the Building
It is easy to talk about all of this in trillions and gigawatts and forget that the Intelligence Grid is being assembled by human beings, and paid for by them too.
The build-out is one of the largest sources of skilled-trade employment in the country right now. Electricians, pipefitters, ironworkers, linemen, concrete crews and HVAC techs are booked years out in the data center corridors. Towns that have not seen major capital investment since the interstate era are suddenly negotiating with hyperscalers. That is real opportunity, the kind that does not require a computer science degree, and it should be counted honestly on the positive side of the ledger.
So should the costs. Data centers already consume a quarter of Virginia’s electricity, and capacity market prices in the mid-Atlantic have surged in ways that show up on ordinary families’ utility bills. Communities are being asked to trade water, land and grid headroom for tax revenue and construction jobs, and not every one of those trades will look wise in ten years. The electrification of America eventually reached every farmhouse, but it took the Rural Electrification Administration and decades of deliberate policy to make sure it did. Abundant intelligence will not distribute itself evenly on its own either. That will be a choice, made by companies, regulators and voters, over and over, for decades.
The Next 250
Which brings me back to Saturday. The founders could not have imagined semiconductors, frontier models or hyperscale data centers. They never saw a light bulb. But they understood something that has held up for two and a half centuries: Infrastructure matters because of what it lets ordinary people do. Roads, ports, post offices and patents were, in large part, machinery for letting individuals pursue their own ambitions, and every great American build-out since has been judged, eventually, by the same standard: Not how much was spent or how impressive the engineering was, but whether it expanded opportunity or concentrated it.
That is the test the Intelligence Grid will face, too. The question for America’s next 250 years is not whether it can build the world’s most powerful intelligence infrastructure. On current evidence, it can and it will. The question is whether that infrastructure ends up preserving the freedom, the competition and the entrepreneurial opportunity that made the first 250 years worth celebrating.
Someday, the photographs of today’s construction sites, the cranes over the Ohio farmland, the cooling towers in the Texas scrub, will hang in the same frame as the pictures of the first power plants, the transcontinental railroad crews and the interstate cloverleafs. Our grandchildren will look at them the way we look at those older photographs now, as the foundations of a prosperity they take completely for granted. Whether they also inherit the freedom to build their own improbable things on top of it, the way every American generation before them did, is being decided right now, one substation at a time. Happy 250th. Now back to work.
