From Orbit to Dyson Swarm: The 100-Year Infrastructure Plan Hidden Inside the SpaceX-Google Deal

From Orbit to Dyson Swarm: The 100-Year Infrastructure Plan Hidden Inside the SpaceX-Google Deal

The Data Center Race Has Left the Atmosphere

Google and SpaceX are reportedly in talks to launch data centers into space. The Wall Street Journal broke the story on May 12, citing sources familiar with the discussions, which center on SpaceX providing rocket launch services as Google moves forward with prototype orbital satellites. The two companies are not simply partnering. They are also competitors: SpaceX has filed with the FCC to launch up to a million satellites for its own orbital data center constellation, a project central to its pitch to investors ahead of a planned IPO valued at $1.75 trillion.

The talks are early, and the technology is unproven at any meaningful scale. But the companies involved, the money behind them, and the hardware already in orbit make this more than a concept pitch.

What Is Actually Happening

Google's orbital initiative, called Project Suncatcher, was announced in late 2024. The goal is to launch a cluster of roughly 81 satellites spanning a one-kilometer radius in low-Earth orbit, carrying the company's custom Tensor Processing Units. Planet Labs is building the satellites. The company plans to launch two prototypes this year, with broader deployment targeted for 2027. Google CEO Sundar Pichai described the ambition in a November interview: "We'll send tiny racks of machines and have them in satellites, test them out, and then start scaling from there."

SpaceX, for its part, has been selling investors on the idea that orbital compute will be cheaper than terrestrial compute within a few years. The company acquired xAI in February in a deal that valued the combined entity at $1.25 trillion. It struck a deal last week with Anthropic to supply 300 megawatts of new computing capacity using more than 220,000 Nvidia GPUs, with Anthropic expressing interest in eventual orbital collaboration. SpaceX is not just a launch provider in this story. It is positioning itself as an AI infrastructure company.

The discussions with Google would make SpaceX the rocket backbone for a competitor's orbital compute ambitions. That kind of arrangement is not unusual in the launch business, where SpaceX dominates commercial access to orbit. Anyone looking to put satellites in space has to at least evaluate a SpaceX deal.

The Case for Space

The argument for orbital data centers starts with power. Terrestrial data centers face real constraints: land acquisition, grid access, water for cooling, permitting, and growing community resistance. In the United States, proposed data center projects have drawn opposition from rural communities concerned about farmland loss, water depletion, energy price increases, and pollution. Orbital facilities sidestep most of those objections.

In space, a satellite in a sun-synchronous orbit can receive near-continuous sunlight. Estimates suggest solar irradiance in those orbits runs roughly 36 percent higher than on Earth's surface, with no night cycle and no weather. Cooling, while not trivial, works through radiative heat dissipation into the cold vacuum, eliminating the need for the massive water cooling systems that make terrestrial data centers so resource-intensive.

The first company to actually put a GPU cluster in orbit was not Google or SpaceX. It was Starcloud, a Y Combinator-backed startup with NVIDIA as a partner, which launched its first satellite carrying an H100-class GPU in late 2025. The company trained a language model in orbit and ran Google's Gemini model on the hardware. That experiment established a proof of concept. The scale was tiny. The principle held.

Axiom Space launched dedicated orbital data center nodes in January 2026. Blue Origin has announced plans under a project called TeraWave. Former Google CEO Eric Schmidt acquired launch company Relativity Space with orbital data centers as an explicit goal. The field has moved from white papers to hardware in roughly 18 months.

The Case Against

The economics are the problem. Terrestrial data centers, even accounting for power costs and land acquisition, remain far cheaper than orbital ones once satellite manufacturing and launch costs are included. Estimates from industry analysts put the cost of a one-gigawatt orbital setup at roughly three times that of an equivalent ground installation. SpaceX's own February 2026 pricing listed standard rideshare rates at $7,000 per kilogram. A single rack of servers is heavy. A constellation of racks is very heavy.

Engineering adds to the cost. Hardware in space is exposed to radiation that degrades electronics faster than any terrestrial environment. Repairs are not possible in any practical sense. Orbital debris is a growing concern. As the number of satellites increases, so does the collision risk, and some researchers have estimated that in a scenario where operators lose control of satellites, cascade collisions in low-Earth orbit could begin within three days.

SpaceX itself disclosed to investors in April that its orbital data center business might not pay off. The company has been transparent about the uncertainty, which is either unusual candor or a hedge against investor claims down the road.

The counterargument to the economics critique is trajectory. Launch costs have fallen dramatically over the past decade, largely because of SpaceX's reusable rocket program. The Starship vehicle, if it achieves the flight cadence SpaceX is targeting, could reduce per-kilogram costs further. The companies betting on orbital compute are betting that the economics will shift, not that they already have.

The Longer Arc

The background research for this piece included a detail worth noting. Futurists and researchers have drawn a direct line from today's orbital data center projects to the concept of a Dyson swarm, a theoretical megastructure consisting of millions or billions of solar-collecting satellites surrounding a star to capture its energy output. The physicist Freeman Dyson proposed the idea in 1960. Sam Altman has referenced "a Dyson sphere of data centers" around the Sun. The CEO of Starcloud has described early Earth-orbit clusters as the seed for a solar-system-scale swarm.

That framing puts the Google-SpaceX talks in a particular light. What looks like a launch services negotiation between two large technology companies is also, in a longer view, the first serious commercial step toward moving computation off the planet entirely. The Starcloud experiment trained a language model on a single GPU in orbit. The next step is a small cluster. Then a constellation. The scale compounds.

Whether that arc bends toward a Dyson swarm or toward a pile of expensive debris depends on engineering problems that have not been solved. Radiation hardening, orbital collision avoidance, and the economics of launch at scale remain open questions. The ambition is real. The execution risk is equally real.

What Comes Next

Google's prototype launches, planned for this year, will test whether machine learning hardware functions reliably in orbit over sustained periods. The results will matter more than any deal announcement. If the hardware degrades faster than expected, or if the latency between ground and orbit proves more limiting than models suggest, the economics get harder.

SpaceX's IPO, anticipated this summer, will surface more detail about its orbital compute financials. The company's pitch to investors has centered on this business line. Public filings will clarify whether the unit economics have improved since the April disclosure to private investors.

For the broader AI infrastructure conversation, the orbital data center race is a direct response to the terrestrial bottlenecks that have slowed data center expansion across the United States and Europe. Land, power, water, and community opposition are real constraints. Space removes most of them, at a cost that remains, for now, prohibitive at scale. The question the next few years will answer is how fast that cost falls.

---

Closing Note

The orbital data center story connects directly to themes I've been developing in my forthcoming book, The Tony Hawk Paradox, which examines how capabilities that appear first in simulated or controlled environments tend to precede their emergence in the physical world. Computing in orbit is following a familiar pattern: proof of concept in miniature, followed by rapid commercial scaling. More on that research at davidborish.com.