Google wants to put AI data centers in space - not in your neighborhood

• Google’s Project Suncatcher envisions clusters of solar-powered satellites packed with the company’s AI chips, effectively creating data centers in orbit.
• The attraction is enormous: nearly continuous solar power, less pressure on local electric grids and water supplies, and fewer fights over where giant terrestrial data centers should be built.
• But the engineering is formidable. Cooling computers in a vacuum, protecting chips from radiation, moving huge quantities of data and avoiding orbital debris could keep large-scale space computing years away.

Google is preparing to test an idea that sounds more like science fiction than cloud computing: moving some of the enormous computing infrastructure needed for artificial intelligence off Earth entirely and away from all the disputes about plopping them down in suburban neighborhoods.

The company’s Project Suncatcher is exploring whether fleets of satellites equipped with Google Tensor Processing Units, or TPUs, could operate as interconnected AI data centers in space, powered largely by sunlight.

Google plans an initial test with two prototype satellites targeted for launch in early 2027 in partnership with satellite operator Planet. The mission is intended to find out whether its AI hardware can survive and operate effectively in orbit, according to blog.google.

The project has gained new relevance as the AI boom collides with a much more earthly problem: data centers consume staggering quantities of electricity, require costly transmission infrastructure and can place heavy demands on water supplies.

Those pressures have increasingly put data centers at the center of fights involving utilities, state regulators and homeowners worried that billions of dollars in new infrastructure will eventually show up in their electric bills.

Space, Google argues, could offer another option.

Why space looks attractive

The basic idea rests on solar energy.

Google researchers say a solar panel in the right orbit can generate as much as eight times the power of the same panel on Earth because it receives far more continuous sunlight and is unaffected by clouds, nighttime and atmospheric losses.

Instead of building a single giant orbital facility, Google envisions constellations of smaller satellites carrying AI processors and communicating with each other using high-speed optical — essentially laser — links.

The satellites would work together as one large computing system.

Google researchers say the approach could eventually provide enormous computing capacity without requiring the land, power plants, transmission lines and water infrastructure demanded by terrestrial data centers.

That is becoming increasingly important as AI computing expands.

The Government Accountability Office said earlier this year that the Department of Energy projects data centers could account for as much as 12% of U.S. electricity demand by 2028, driven largely by artificial intelligence.

Google itself acknowledged the problem in describing its terrestrial infrastructure strategy, noting that AI computing demand is increasingly outrunning the space and electrical capacity of individual data centers.

A possible escape valve for the grid

For consumers, the intriguing part of the Suncatcher experiment has little to do with rockets.

It is what happens to the electric grid.

Data-center expansion has forced utilities across the country to consider new power plants, substations and transmission lines. Who ultimately pays for that infrastructure has become a major regulatory question.

Some utilities and states have created special tariffs requiring data-center operators to shoulder more of their costs. Others allow portions of infrastructure spending to be spread among customers.

Moving even part of future AI computing into orbit could eventually reduce those pressures.

That does not mean neighborhood data centers are about to disappear.

Even advocates of orbital computing generally expect Earth-based facilities to remain essential, particularly for applications that require near-instant responses. Space computing could initially be better suited for enormous AI-training jobs that can tolerate somewhat greater communication delays.

There is a very large catch: heat

Solar power may be abundant in space.

Cooling is not.

Computers generate tremendous amounts of heat, and although space is extremely cold, a vacuum does not carry heat away the way air or water does.

An orbital data center would therefore need large radiators capable of releasing heat through radiation.

The GAO calls heat disposal one of the central engineering problems facing space-based data centers. Large-scale cooling technology of the kind required has not yet been demonstrated.

Radiation poses another problem.

High-energy particles can corrupt data or damage electronic components. Google says it has already subjected its TPUs to radiation testing and found encouraging results, but real orbital conditions will provide a much more demanding test.

Then there is the cost of getting everything up there

Launching tons of computers, solar arrays, radiators and communications equipment remains extremely expensive.

The economic case for orbital data centers therefore depends heavily on reusable rockets continuing to push launch costs downward.

One industry analysis estimates that space computing could become substantially more competitive if launch costs fall below roughly $500 per kilogram. Current Falcon 9 costs have been estimated at several times that amount, while next-generation reusable spacecraft such as SpaceX's Starship are intended to push costs dramatically lower.

Google has also held discussions with SpaceX and other potential partners as it explores how a larger Suncatcher system could ultimately be launched, according to Reuters.

Google is hardly alone.

SpaceX, Blue Origin and several startups are pursuing various forms of orbital computing infrastructure, turning what recently seemed an eccentric idea into a developing technology race.

And space is already getting crowded

There is another environmental question.

Putting thousands — or potentially hundreds of thousands — of computing satellites into orbit could worsen the growing problem of space debris.

More spacecraft increase the chances of collisions, which can generate clouds of fragments capable of striking still more satellites.

The GAO says large orbital data centers could also complicate astronomical research, radio-frequency allocation and space-traffic management.

That creates an odd tradeoff.

Orbital data centers could reduce the environmental and infrastructure burden imposed on communities on Earth while creating new environmental and regulatory problems above it.

What happens next

The first major test comes in 2027.

Google’s two prototype Suncatcher satellites are not intended to replace a terrestrial data center. They are engineering experiments designed to find out whether powerful AI processors can reliably operate, communicate and survive in orbit.

If they work, substantially larger experiments are likely to follow.

If they don't, engineers will at least learn which parts of the concept need to change.

Either way, the project illustrates how extraordinary the AI infrastructure race has become.

Companies that once debated where to put the next data center are increasingly asking a much bigger question:

Does the next one need to be on Earth at all?

What this means for consumers

The immediate impact is minimal; Google is still testing the technology.

But orbital computing could eventually matter to households because the biggest constraint on AI growth is increasingly electricity.

If AI companies can obtain substantial computing power without demanding new terrestrial power plants and transmission systems, fewer of those infrastructure costs may wind up being fought over in state utility proceedings — and potentially passed along to residential customers.

There is plenty of uncertainty. Space-based computing could prove too expensive or technically difficult to operate at large scale.

But Google’s experiment suggests the industry is beginning to recognize something consumers living near rapidly expanding data-center regions already know:

The AI boom ultimately has to get its electricity from somewhere.

--

ChatGPT provided research for this article.