America’s AI Revolution Will Demand Much More Space

November 19, 2026

You’re reading Dispatch Markets, a weekly dive into the forces driving economic growth—and those holding it back—featuring Scott Lincicome, Kyla Scanlon, Karl Smith, Marian Tupy, and Adam Ozimek.


On November 3, voters sent a clear message to their elected representatives. Even though Republicans experienced a difficult night, the data-center sector might have envied them. Ballotpedia tracked 53 local contests, and in 52 of them voters favored curbs on data centers or rejected proposals that would have made construction easier.

What’s more, the sole victory carried a caveat: in Harwood, North Dakota, residents rejected an amendment that would have demanded additional review for manufacturing, warehousing, and energy projects, including data centers. Measures aimed specifically at data centers fared worse for AI advocates. In Sunbury, Ohio, 79 percent supported banning data centers above 25 megawatts, and in Edgewater, Florida, 74 percent backed a ban on new data centers altogether.

Estimates from Lawrence Berkeley National Laboratory suggest data centers used about 4.7 percent of U.S. electricity in 2024, with a potential rise to nearly 12 percent by 2030. A natural question is who pays for all that power. Virginia’s legislative auditors reported that while data centers cover the utility costs allocated to them, the infrastructure required to meet growing demand could raise bills for everyone else.

Water-use concerns were equally pressing. Even as the industry touts newer cooling approaches as more efficient, voters remained wary about potential impacts on local drinking-water supplies. If developers could address every electricity bill concern and reassure communities about water, would that be enough? Most observers doubt it.

Researchers at the University of Virginia asked 809 people to imagine one of four facility types tucked within ten miles of their homes. A data center aroused more opposition than even a nuclear plant. Among respondents who believed the developer would take advantage of the community, 90 percent withheld support. Alarmingly for those pursuing America’s AI ambitions, the survey found that even 41 percent of respondents who defaulted to giving the developer the benefit of the doubt still opposed the project.

This leaves unresolved whether distrust leads to opposition or vice versa. It also gives reason to seriously weigh the sense that residents feel decisions are being made without their input. A recent survey found that 62 percent believed officials would give residents little or no consideration when deciding where and whether to host data centers in their area.

All of this comes at a moment when workers are being told that AI will soon reshape the kinds of tasks they’ve spent their lives performing. They often have little say over how or when those changes will occur, and they don’t know how profoundly their livelihoods may be disrupted.

Then a local ballot question appears about whether an AI company can construct a data center in their town. This could be their only real chance to push back against the tide. It’s no surprise voters reacted as they did, and modest improvements to water cooling or on-site power generation by data centers are unlikely to alter opinions. If America wishes to retain its edge in building data centers, the challenge is to find places to host them.

Fortunately, one last frontier remains: space. Google is testing AI chips in low Earth orbit, and SpaceX has placed a bold bet on orbital data centers.

To many, this may sound far-fetched. I was skeptical at first, too, but repeated cost-and-feasibility checks convinced me otherwise. With a sufficiently affordable launch and abundant sunlight, the math of placing servers on satellites in space—rather than in vast buildings nearby—begins to look feasible.

 Outer space is closer than it used to be. 

Space is immense. The nearest region, known as low Earth orbit, stretches roughly 60 to 1,240 miles above the planet. For decades, satellite operators favored a more distant zone called geostationary orbit. In geostationary orbit, a satellite circles about 22,236 miles above the equator in the same direction as the Earth’s rotation. One orbit takes 23 hours and 56 minutes, roughly matching the planet’s spin, making the satellite appear fixed in the sky.

Geostationary orbit is convenient for transmitting TV signals to a specific region or for weather satellites that monitor a single swath of Earth. However, the higher the orbit, the more challenging the ascent.

The space shuttle Challenger era saw NASA place the first tracking and data relay satellite into orbit in 1983. Yet Challenger could only reach about 200–300 miles up, the lower edge of low Earth orbit. From there, a separate rocket carried the satellite higher, but a malfunction forced the craft to climb via its own tiny thrusters for months to reach its destination.

That approach was clumsy, yet the alternative would have been even more prohibitive: to reach geostationary orbit in a single launch, you’d need a rocket roughly two-thirds as large as Saturn V, and nothing would be reusable. Building such a colossal rocket for every mission made satellites a rarity, and space ventures became the realm of government and scientific agencies.

The space shuttle concept aimed for reusability but ended up as a product of bureaucratic design, with limited range and reuse. The enormous orange external tank was discarded on every flight, the two solid-fuel boosters were expensive to recover and refurbish, and the orbiter’s memory-elite heat shield depended on thousands of bespoke tiles assessed by hand after each mission.

Fortunately, market forces came to the rescue. SpaceX’s Falcon Heavy can reach geostationary orbit and return two of its three boosters to Earth. More importantly, SpaceX’s workhorse, the Falcon 9, can deliver a payload to low Earth orbit and return intact. The internet is full of clips showing the rocket landing back on its engines, tail-first in a feat of reusable flight.

A 2018 NASA study compared shuttle-era costs to get to low Earth orbit with modern rockets, finding about $61,700 per kilogram for the shuttle versus $1,410 per kilogram for the Falcon Heavy. That roughly 40-fold difference has transformed the economics of space deployment.

Introduction to orbital economics.

Midcentury America brimmed with visions of a space-born economy, where life and work could extend above the clouds, potentially up to the lunar surface. Yet progress has not followed those early fantasies; technology advanced more gradually than imagined.

What remains true, however, is that escaping Earth’s gravity is arduous, and for most ventures the advantages of space are limited. A few cases justify the effort: those who want wide-area Earth observation, whether for weather, defense, or connecting remote regions. In those instances, venturing toward geostationary orbit can be worthwhile, even if it means going to substantial altitudes.

Low Earth orbit offers proximity, but it introduces two perennial challenges: you must deploy more satellites to maintain continuous coverage, and they will need replacement far sooner than GEO satellites.

Put simply, the lower the orbit, the quicker a satellite circles the globe: the circumference is smaller, and gravity tugging on it is stronger. A GEO satellite completes one circuit daily, whereas a low-Earth satellite zips around roughly every 90 minutes.

Because a GEO satellite remains fixed relative to the Earth’s rotation, it looks stationary in the sky. A low-Earth satellite, by contrast, lingers over a given area for only about ten minutes before fading from view as it moves along the horizon.

If uninterrupted coverage over a single region is desired, the ratio works out to nine low-Earth orbit satellites for every GEO satellite.

On top of that, while the atmosphere in low-Earth orbit is extremely thin, it still matters. Drag causes low-Earth satellites to descend after roughly five years, whereas GEO satellites can persist for decades, though their internal systems typically fail after about thirty years. All told, maintaining continuous coverage requires far more low-Earth satellites, making space-based constellations costly to sustain—before reusable rockets slashed launch expenses.

It’s always sunny in space.

Naturally, once a satellite is in orbit, fuel efficiency becomes crucial. Over the last twenty years, the cost of solar power has fallen about 90 percent, from roughly 41 cents to around 4 cents per kilowatt-hour. By comparison, electricity generated by natural gas plants runs around 9 cents per kilowatt-hour. That makes solar the cheapest source of power—yet sunlight in space is constant, which changes the calculus. A data center requires around-the-clock operation, so a solar-powered setup would need storage or a backup source, which cuts into savings. In space, however, the sun never truly sets—if a rover or satellite can trace the line between day and night across the globe, it can stay in perpetual daylight, avoiding the need for batteries.

Moreover, sunlight in space is more intense, and the absence of weather eliminates the need for protective glass shielding for sensitive electronics. All told, those conditions can trim power costs by roughly 90 percent.

Hot chip, no lie.

A central challenge in operating a data center is cooling the silicon brains. AI chips generate a lot of heat and are vulnerable to performance losses when temperatures rise above 100 Celsius (212 Fahrenheit), and they can fail entirely beyond about 120 C.

On Earth, cooling usually means circulating cold water through narrow tubes linked to the servers, transferring heat into the water, which then warms up. The used water must go somewhere, often evaporating in the past.

That approach consumes enormous water and has long troubled environmentalists. Newer cooling methods exist, but they tend to cost more.

In space, cooling relies on infrared radiators. Liquid coolant draws heat from the chips into broad flat panels, which radiate infrared light into the cold void. The catch is governed by the Stefan-Boltzmann law: for a given temperature rise, the radiated energy climbs with the fourth power of that rise. In practice, this means hotter chips translate into dramatically cheaper cooling in space.

Pencils out.

Let’s attach a price tag to the idea. Elon Musk’s target remains a launch cost of $10–$15 per kilogram, though his figures have sometimes been optimistic. Conservative estimates place SpaceX’s eventual cost around $200 per kilogram. A megawatt of computers, solar panels, radiators, and supporting gear to orbit would run about $7 million.

It might look steep at first glance, but compare that to a terrestrial data center’s bill. Building a facility with solar power, battery backup, and water-sparing cooling runs just under $29 million per megawatt—not counting taxes or land. By contrast, orbital solar panels, infrared radiators, and the rest of the space hardware total around $4 million per megawatt.  

The engineering is steadily maturing, though large-scale economics remain to be proven. Still, at about $200 per kilogram, the math adds up. And judging by recent election results, getting a data center into orbit could eventually be easier than navigating a local zoning board.

Markets FTW

We’ve all witnessed the remarkable feats achieved by humanoid robots crafted by firms such as Boston Dynamics. Yet it should come as no surprise to students of capitalism that one of the earliest commercial uses of so-called physical AI sits in agriculture.

Traditional industrial robots perform a single, highly precise task repeatedly. They’re efficient but lack flexibility. Physical AI couples the adaptability and knowledge found in large language models like ChatGPT with autonomous machinery.

Carbon Robotics has developed a system capable of identifying weeds and precisely firing a laser at the weed’s critical meristematic tissue—the part necessary for growth. Laser weeding can reduce, and in some cases replace, the need for herbicides.

Herbicides pose health concerns for many consumers and also cut yields because no chemical is perfectly selective for weeds without some crop risk.

What’s notable is that this robot is designed as a standard tractor attachment, making the switch to laser weeding nearly seamless for farmers. Looking ahead, Carbon Robotics plans to offer a kit to convert a farmer’s tractor into a self-driving unit. The pastoral calm of farm life has never looked as appealing.

Chart of the Week

The cost trajectory of sending hardware into space using the most advanced rockets over time. (Source: Visual Capitalist)


Disclaimer: The opinions expressed above do not necessarily reflect those of the presenting sponsor.

Pilar Marrero

Political reporting is approached with a strong interest in power, institutions, and the decisions that shape public life. Coverage focuses on U.S. and international politics, with clear, readable analysis of the events that influence the global conversation. Particular attention is given to the links between local developments and worldwide political shifts.