Google lately unveiled Undertaking Suncatcher, a analysis “moonshot” aiming to construct a information middle in house. The tech large plans to make use of a constellation of solar-powered satellites which might run by itself TPU chips and transmit information to at least one one other through lasers.
Google’s TPU chips (tensor processing models), that are specifically designed for machine studying, are already powering Google’s newest AI mannequin, Gemini 3. Undertaking Suncatcher will discover whether or not they could be tailored to outlive radiation and temperature extremes and function reliably in orbit. It goals to deploy two prototype satellites into low Earth orbit, some 400 miles above the Earth, in early 2027.
Google’s rivals are additionally exploring space-based computing. Elon Musk has mentioned that SpaceX “will likely be doing information facilities in house,” suggesting that the following era of Starlink satellites could possibly be scaled as much as host such processing. A number of smaller corporations, together with a US startup known as Starcloud, have additionally introduced plans to launch satellites outfitted with the GPU chips (graphics processing models) which are utilized in most AI techniques.
The logic of knowledge facilities in house is that they keep away from lots of the points with their Earth-based equivalents, notably round energy and cooling. Area techniques have a a lot decrease environmental footprint, and it’s probably simpler to make them greater.
As Google CEO Sundar Pichai has mentioned: “We’ll ship tiny, tiny racks of machines and have them in satellites, take a look at them out, after which begin scaling from there … There isn’t any doubt to me that, a decade or so away, we will likely be viewing it as a extra regular strategy to construct information facilities.”
Assuming Google does handle to launch a prototype in 2027, will it merely be a high-stakes technical experiment—or the dawning of a brand new period?
The Scale of the Problem
I wrote an article for The Dialog at the beginning of 2025 laying out the challenges of placing information facilities into house, through which I used to be cautious about them occurring quickly.
Now, after all, Undertaking Suncatcher represents a concrete program quite than simply an concept. This readability, with an outlined aim, launch date, and {hardware}, marks a major shift.
The satellites’ orbits will likely be “solar synchronous,” which means they’ll at all times be flying over locations at sundown or dawn in order that they’ll seize daylight practically constantly. Based on Google, photo voltaic arrays in such orbits can generate considerably extra power per panel than typical installations on Earth as a result of they keep away from dropping daylight attributable to clouds and the ambiance, in addition to at night time.
The TPU assessments will likely be fascinating. Whereas {hardware} designed for house usually must be closely shielded towards radiation and excessive temperatures, Google is utilizing the identical chips utilized in its Earth information facilities.
The firm has already executed laboratory assessments exposing the chips to radiation from a proton beam that counsel they’ll tolerate nearly 3 times the dose they’ll obtain in house. That is very promising, however sustaining dependable efficiency for years, amidst photo voltaic storms, particles, and temperature swings is a far more durable take a look at.
One other problem lies in thermal administration. On Earth, servers are cooled with air or water. In house, there isn’t any air and no easy strategy to dissipate warmth. All warmth should be eliminated via radiators, which regularly turn out to be among the many largest and heaviest elements of a spacecraft.
NASA research present that radiators can account for greater than 40 % of whole energy system mass at excessive energy ranges. Designing a compact system that may maintain dense AI {hardware} inside secure temperatures is likely one of the most troublesome facets of the Suncatcher idea.
An area-based information middle should additionally replicate the excessive bandwidth, low latency community cloth of terrestrial information facilities. If Google’s proposed laser communication system (optical networking) goes to work on the multi-terabit capability required, there are main engineering hurdles concerned.
These embody sustaining the required alignment between fast-moving satellites and dealing with orbital drift, the place satellites transfer out of their meant orbit. The satellites can even need to maintain dependable floor hyperlinks again on Earth and ovecome climate disruptions. If an area data-center is to be viable for the long run, it is going to be important that it avoids early failures.
Upkeep is one other unresolved difficulty. Terrestrial information facilities depend on continuous {hardware} servicing and upgrades. In orbit, repairs would require robotic servicing or further missions, each of that are expensive and sophisticated.
Then there’s the uncertainty round economics. Area-based computing turns into viable solely at scale, and provided that launch prices fall considerably. Google’s Undertaking Suncatcher paper means that launch prices may drop beneath $200 (£151) per kilogram by the mid 2030s, seven or eight instances cheaper than at this time. That may put development prices on par with some equal services on Earth. But when satellites require early substitute or if radiation shortens their lifespan, the numbers may look fairly totally different.
Briefly, a two-satellite take a look at mission by 2027 sounds believable. It may validate whether or not TPUs survive radiation and thermal stress, whether or not solar energy is secure, and whether or not the laser communication system performs as anticipated.
Nevertheless, even a profitable demonstration would solely be step one. It might not present that large-scale orbital information facilities are possible. Full-scale techniques would require fixing all of the challenges outlined above. If adoption happens in any respect, it’s prone to unfold over many years.
For now, space-based computing stays what Google itself calls it, a moonshot: bold and technically demanding, however one that would reshape the way forward for AI infrastructure, to not point out our relationship with the cosmos round us.
This text is republished from The Dialog beneath a Inventive Commons license. Learn the authentic article.
