Nvidia Space-1 Orbital AI Data Centers: Who's Racing to Put Chips in Space 2026

Nvidia just announced an AI chip designed to run in orbit, and at least five companies are racing to put data centers in space — while one of the world's top analyst firms calls the entire idea "peak insanity."

The Vera Rubin Space-1 Module, unveiled at Nvidia's GTC 2026 conference on March 16, is purpose-built for satellites and orbital data centers. It packs 25 times the AI compute power of an H100 GPU into a package engineered for the extreme constraints of space: limited size, weight, power, and zero conventional cooling.

"Space computing, the final frontier, has arrived," Nvidia CEO Jensen Huang told the audience. "As we deploy satellite constellations and explore deeper into space, intelligence must live wherever data is generated."

That's a bold claim. But Nvidia isn't alone in betting on it.

The race to orbit

The list of companies trying to build data centers above the atmosphere has grown fast.

SpaceX filed with the Federal Communications Commission in January 2026 to launch up to one million satellites for AI data centers. The FCC fast-tracked the application — no environmental impact assessment required — a decision that has alarmed astronomers who say the constellation could ruin ground-based observation.

Google announced Project Suncatcher in late 2025, partnering with Planet Labs to explore arrays of solar-powered satellites equipped with the company's TPU AI chips. Two prototype satellites are expected by early 2027, with the long-term vision being 81-satellite clusters arranged in one-kilometre formations.

Starcloud, backed by Nvidia, has already trained its first AI model in space and plans to deploy a fleet of 88,000 satellites carrying AWS Outposts hardware. Aetherflux aims to launch its first data center satellite into orbit in the first quarter of 2027. Axiom Space, Kepler Communications, and Sophia Space are all building on Nvidia's space computing platforms.

And then there's Elon Musk. SpaceX acquired xAI in February 2026 in a $1.25 trillion merger — the largest in history — explicitly aimed at building AI infrastructure in orbit. Internal valuations for the combined entity have since climbed to $1.75 trillion as markets reprice SpaceX as an AI infrastructure company, not just a rocket launcher.

Why space? Because Earth is running out of power

The logic driving this rush isn't about glamour. It's about electricity.

AI data centres are devouring power at a pace that terrestrial grids can't sustain. The buildout has been blamed for soaring electricity costs across the US, with utilities and ratepayer advocates pushing back hard. Communities that welcomed data centres for jobs are now fighting them over water consumption and grid strain.

Space, by contrast, offers virtually unlimited solar energy with no neighbours to complain. And for certain workloads — geospatial intelligence, satellite imagery processing, autonomous spacecraft operations — processing data where it's collected eliminates the latency and bandwidth costs of beaming everything back to Earth.

The Nvidia Rubin Module isn't designed to replace ground data centres. It's meant for what the company calls "space-based inferencing": running AI models on data gathered by orbital sensors. Think real-time analysis of satellite imagery, autonomous navigation for spacecraft, and intelligence processing that happens 400 kilometres up instead of on the ground.

The sceptics aren't quiet

Not everyone thinks this makes sense.

Gartner, the technology research firm, published a report in February titled with characteristic directness. Distinguished VP analyst Bill Ray called orbital data centres "peak insanity."

"Companies are wasting money by pouring funds into the orbital data centre 'bubble' because the economics do not work," Ray wrote. "This is due to the prohibitive costs of launching hardware and the immense technical challenges of cooling these orbital data centres in the vacuum that is space."

That cooling problem is real. On Earth, data centres use air conditioning, water cooling, or liquid immersion to manage heat. In space, there's no air to carry heat away — only radiation, the slowest form of thermal transfer. Jensen Huang acknowledged this directly: "In space, there's no convection, there's just radiation, and so we have to figure out how to cool these systems out in space."

Ray went further, warning that the hype around orbital data centres could actually hurt Earth-based infrastructure: "Product leaders face the very real possibility of underbuilt terrestrial data centre capacity if this 'bubble' of space data centre hype lasts for several years."

Even Sam Altman, OpenAI's CEO, has reportedly described some orbital data centre plans as "ridiculous" and "AI snake oil."

The numbers that matter

Here's what makes this story genuinely strange. The companies investing in orbital compute include some of the most serious names in technology. This isn't a collection of startups with pitch decks and nothing else.

Nvidia's market value exceeds $3 trillion. The SpaceX-xAI merger was valued at $1.25 trillion. Google's Project Suncatcher comes backed by one of the world's deepest R&D budgets.

These are companies that can afford to be wrong — and that's exactly the point. Huang has been explicit about it: launching data centres into orbit is a "poor economic decision" right now. But he'd rather be ready for a boom that never comes than miss one that does.

The question is whether "ready" means a few experimental modules or a million satellites filling low-Earth orbit. SpaceX's FCC filing is for a constellation that would dwarf anything humanity has ever launched. Scientists have already warned it could generate enough light pollution to seriously degrade astronomical observation — and the application was approved without any environmental review.

What this actually means

The orbital data centre race tells us something important about where AI is headed. The industry isn't just growing — it's growing so fast that it's literally running out of planet.

When the power grid can't keep up, when communities revolt against new construction, when water tables drop from cooling demands, the response from the world's richest companies isn't to slow down. It's to leave Earth entirely.

Whether that's visionary or absurd probably depends on which decade you're asking from. The first undersea cables seemed like madness too. So did the idea of connecting a million computers into a global network.

But there's a difference between laying cable across the ocean floor and cooling a GPU in the vacuum of space. The engineering challenges are real, the economics are currently terrible, and the environmental questions — from orbital debris to light pollution to the sheer resource cost of launching hardware — haven't been seriously addressed.

For now, Nvidia's Space-1 module is real hardware solving a real problem: processing orbital data in orbit instead of beaming it down first. That's practical and genuinely useful.

The million-satellite constellation? That's a different bet entirely. And the gap between those two visions — one a useful tool, the other a possible revolution or a spectacular waste — is where the most interesting story in AI infrastructure is playing out right now.