SPACE-BOUND: Musk Predicts Computing Will Move Almost Entirely to Space as Google Prepares Orbital AI Test
Elon Musk is predicting that the future of computing lies overwhelmingly beyond Earth, declaring that virtually all computing capacity will eventually migrate into space as Google moves ahead with an ambitious project to test artificial-intelligence hardware in orbit.
“The amount of compute in space will obviously round up to 100% of all compute,” Musk wrote Thursday on X, responding to a post about Google’s Project Suncatcher, a research initiative aimed at determining whether large-scale AI computing infrastructure could eventually operate aboard networks of satellites.
The prediction is extraordinary in scope. Today, the enormous computing resources needed to train and operate advanced AI systems are concentrated primarily in terrestrial data centers, facilities that require vast quantities of electricity and significant cooling infrastructure. Musk is suggesting that, over the long term, those economics could shift so dramatically that Earth-based computing would represent only a tiny fraction of the world’s total computing power.
Google is now taking an early step toward determining whether such a future is technically feasible.
Through Project Suncatcher, Google is investigating whether satellites carrying its specialized Tensor Processing Units, or TPUs, could function as interconnected AI data centers in space. Google describes the effort as a “research moonshot” and stresses that it remains at an experimental stage.
The concept centers on compact groups of satellites operating in low-Earth orbit. Each satellite would carry Google’s AI processors, while high-speed optical links using lasers would allow the spacecraft to communicate with one another. Rather than having one enormous computer in orbit, multiple satellites could potentially combine their computing resources to tackle larger AI workloads.
One of the biggest attractions is energy.
Data centers on Earth are facing rapidly increasing electricity requirements as AI models grow larger and more computationally demanding. In orbit, Google says solar panels can potentially generate as much as eight times the power they would produce on Earth, depending on the orbit, while receiving sunlight almost continuously. That could reduce both the need for batteries and reliance on terrestrial power grids.
The amount of available energy is enormous. Google researchers noted when unveiling Project Suncatcher that the sun emits more than 100 trillion times humanity’s total electricity production. The company is investigating whether some of that effectively continuous solar power could eventually support massive computing networks in orbit.
But getting an AI data center into space involves far more than attaching computer chips to satellites.
Google researchers say the spacecraft would have to fly in extremely precise formations so that enormous quantities of data could move rapidly between them. The company plans to use free-space optical communications — essentially laser connections — rather than conventional radio links.
The precision required is formidable. Google says the satellites would need to maintain extremely high-bandwidth laser connections while both spacecraft are moving at orbital speeds. The company compared the challenge to attempting to hit a coin-sized target from miles away while both the transmitter and target are moving.
There is also the question of whether sophisticated AI processors can survive the harsh conditions of space.
Google has already subjected its Trillium-generation TPUs to radiation testing designed to simulate conditions in low-Earth orbit. According to Google CEO Sundar Pichai, the chips survived those tests without damage. The company cautions, however, that radiation is only one challenge; thermal management, long-term reliability and the physical stresses associated with launch and orbital operation must also be addressed.
Google is now preparing to move those experiments from the laboratory into space.
The company said Thursday that Project Suncatcher is preparing an initial orbital test aboard SpaceX’s Transporter-18 rideshare mission. The prototype spacecraft was developed in partnership with satellite company Planet and is intended to collect real-world data on how Google’s TPUs withstand radiation, temperature extremes and other conditions in orbit.
That initial experiment will be followed by a more ambitious milestone. Google and Planet plan to put two prototype satellites into orbit in 2027, allowing researchers to test whether two AI-equipped spacecraft can fly together and exchange enormous amounts of data through high-bandwidth optical links. Planet originally announced that the two-satellite mission was targeting launch by early 2027.
The involvement of SpaceX creates an unusual intersection between Musk’s prediction and Google’s experiment: the company whose orbital-AI project prompted Musk’s sweeping forecast is using Musk’s rocket company to help test the concept.
Musk’s statement came in response to an X post claiming that Google was “launching TPUs in space NEXT WEEK on SpaceX falcon 9 to test AI data centers orbit.” Google’s latest announcement does confirm that its first Suncatcher test will fly aboard SpaceX’s Transporter-18 mission, although the company’s own description presents the flight as an initial hardware experiment rather than the deployment of an operational orbital AI data center.
Even if those tests succeed, enormous technical and economic obstacles would remain before anything resembling a large-scale space data center could become practical.
Hardware failures that are routine inconveniences in terrestrial data centers become much harder to address when equipment is hundreds of miles above Earth. Heat must be dissipated in a vacuum, satellites must survive radiation for extended periods, and thousands of processors would need exceptionally fast communications links. Launch costs would also have to become sufficiently low to make replacing and expanding orbital computing hardware economically practical.
Google itself is careful not to portray Project Suncatcher as a near-term replacement for conventional data centers. The company describes the effort as a long-range experiment intended to answer fundamental engineering questions and says the first mission is primarily about discovering what works, what fails and what needs to be redesigned.
Still, Musk’s prediction takes the idea to its ultimate conclusion. Rather than envisioning space-based computing as a supplement to Earth’s data centers, he is forecasting a future in which the overwhelming majority of computing happens in orbit.
“The amount of compute in space will obviously round up to 100% of all compute,” Musk wrote.
For now, Google is considerably closer to the beginning of that experiment than to Musk’s envisioned endpoint. But with actual AI processors preparing to head into orbit and two interconnected prototype satellites planned for 2027, the idea of an AI data center in space is moving from theoretical research toward real-world testing.
{Matzav.com}
