SpaceX and Google want data centers in orbit?

The Wall Street Journal revealed on May 12, 2026: SpaceX and Google are reportedly in advanced negotiations to deploy data centers… in Earth’s orbit. Neither company has officially confirmed the information. And yet, it’s hard to simply wave it away.

Putting servers in space? It sounds crazy at first glance. Until you look at the numbers. And then the idea starts to look less like science fiction, and more like a serious response to a very concrete problem.

Here’s why this prospect, as spectacular as it may be, is becoming credible and achievable.

Why AI is suffocating terrestrial data centers

The starting point is a silent but massive energy crisis. In 2024, data centers consumed 415 TWh of electricity worldwide, according to the International Energy Agency. The projection for 2030? 945 TWh. That’s an increase of +128% in six years.

The 1,000 TWh threshold could even be crossed as early as 2026 — equivalent to Japan’s total electricity consumption. To power increasingly hungry AI models, dedicated servers are growing at a rate of +30% per year.

And the problem is that Earth presents three physical walls to this growth:

• Electricity: grids are saturating, costs are exploding, connection lead times are growing longer
• Land: finding available land close to a reliable energy source is becoming a quest
• Water: cooling systems consume millions of liters every day

This is no longer a growth problem to manage. We’re talking about a physical wall. This is exactly what the massive investments in AI data centers illustrate, such as xAI’s in Memphis — titanic projects at staggering costs, just to scrape together a few extra megawatts.

Orbit as a radical solution: the concrete advantages

Space breaks all three of these constraints at once. Not metaphorically — physically.

In orbit, solar panels capture 5 times more energy than on the ground. No night, no atmosphere to filter and attenuate. Continuous, stable, predictable output.

For cooling, the spatial vacuum at −270 °C does the work for free. No more energy-hungry air conditioning, no more water basins. The savings are massive.

And then, zero terrestrial regulatory constraints. No building permits. No neighbors. No national power grid to negotiate with for years.

The proof of concept already exists: at the end of 2025, Starcloud became the first company to train an LLM directly in orbit. This is no longer a theoretical hypothesis.

Radically advantageous on paper, then. The obstacles come a little further down the road.

What SpaceX brings to the table

SpaceX isn’t coming to these negotiations empty-handed. Far from it.

The company has filed a request with the FCC for 1 million satellites at altitudes ranging from 500 to 2,000 km. This is not a rumor — it’s an official, public, consultable document. These satellites would carry a projected AI computing capacity of 100 gigawatts.

SpaceX also has three unique assets:

• The already-deployed Starlink network, with its operational inter-satellite communications
• Starship, the only launch vehicle capable of delivering massive payloads at a potentially viable cost
• Vertical mastery of the entire chain — from launcher to constellation

Elon Musk stated in early 2026 that space would become the most profitable place for AI within “30 to 36 months”. His timelines are often optimistic — that’s an established fact. But SpaceX’s request for one million satellites dedicated to data centers is real and documented.

Google’s secret project: Suncatcher

On Google’s side, the WSJ mentions the Suncatcher project — an internal initiative, a code name, not yet an announced product. No official confirmation at this stage.

The plan as described: two prototype satellites equipped with TPUs — Google’s in-house AI chips — scheduled for early 2027. Sundar Pichai is said to have internally discussed the vision of a “normal orbital data center” within ten years.

TPUs in orbit: an innovation within an innovation

A TPU (Tensor Processing Unit) is a chip designed specifically for machine learning computations. Google has been equipping its own data centers with them for years — it’s their competitive advantage in silicon.

The problem? Space is a hostile radiation environment. Cosmic particles degrade consumer electronics within weeks. Getting a TPU to survive and perform in orbit is an innovation within an innovation.

If Google succeeds with this prototype, it’s not just a victory for Google. It’s a paradigm shift for the entire space semiconductor industry.

Technical obstacles not to be underestimated

Let’s be clear-eyed. Between the idea and an operational infrastructure, there is a list of obstacles that are anything but symbolic.

• Radiation-resistant hardware: the entire current chip ecosystem needs to be rethought to survive in orbit
• Inter-satellite laser communications: transferring petabytes from space with acceptable latency remains an unsolved challenge at scale
• Orbital debris management: one million additional satellites in an already crowded space is a real systemic risk — Kessler syndrome is not a metaphor
• Launch economics: even with Starship, the cost per kilogram remains a major limiting factor for multi-ton infrastructures
• Competition: Amazon and Blue Origin are working on similar projects — this race has multiple runners

To power all of this, terrestrial energy solutions are already showing their limits. Projects like Tesla Megapack-based data center power solutions illustrate just how much every kilowatt-hour counts — and why orbit, with its permanent sunlight, makes engineers dream.

The question is no longer really whether orbital data centers will exist. The signals — FCC filings, prototypes, investments — are converging too clearly. The real question is when. And today’s obstacles are precisely tomorrow’s roadmaps.

As the experts consulted by Le Monde confirm: by 2030, data center consumption will reach the equivalent of Japan’s total consumption. At that level of pressure, even the most radical solutions become reasonable.