Larry Page reportedly has a number for how much of the world's computing will eventually sit in orbit, and Elon Musk thinks he's right. The two haven't spoken in years, so the endorsement is worth a closer look. Coming from two people who fell out long ago, a public nod like this doesn't happen by accident, and it's fair to ask what each side gets out of it.
The prediction and the endorsement
The claim didn't come from a press release. It surfaced on the Nov. 4 episode of Invest Like the Best, where host Patrick O'Shaughnessy recalled that Page was asked how much of the world's computing power, meaning the servers that run AI and everything else online, would eventually be in space. Page's reported answer was "five nines," or 99.999%.
That phrase is an engineering term for availability, meaning a system that's down for only a few minutes a year. Using it for orbital computing is a sharp choice. Podcast answers are loose and unscripted, and it's not clear from the account whether Page meant a precise bet or a rhetorical flourish. Still, sharp phrases travel, and this one has.
Musk shared the clip on X on Sunday, and he kept the commentary short. The verdict fits in a single line: "Larry Page has made many accurate predictions and this is one of them." Brevity is its own signal here. A man who posts constantly chose not to elaborate, which makes the statement read more like a stamp of approval than an argument.
Read that as a pitch as much as a compliment. SpaceX absorbed Musk's AI company xAI this year and wants to build AI data centers in space. Earlier this month, Musk argued that AI hardware could be easier to cool in orbit than on Earth. That's the core bet: if cooling is a constraint on the ground, orbit starts to look like a sensible place to put the racks.
Here's the part worth pausing on. Terrestrial data centers spend real money moving heat away from hardware, and cooling is a meaningful share of their operating burden. Space has its own thermal problem, though. With no air to carry heat away, a spacecraft has to radiate it, which is a different engineering challenge rather than an obviously easier one. Musk's argument rests on a trade-off, not a free lunch.
My read is the endorsement helps SpaceX's roadmap at least as much as it validates Page's forecast.
Google is running the same experiment
Musk isn't the only one testing the idea. Google is exploring it through Project Suncatcher, and its chips recently generated their first AI response in orbit. That's a small technical milestone, but it's the kind of demonstration that moves an idea from concept toward hardware planning.
Running a single inference in orbit is a long way from running a full training cluster there. One response proves the hardware can function and the software can talk to it. It doesn't prove the economics work, and the economics are where these projects tend to stall. If you're working in infrastructure, that gap is where the hard questions live.
An old rivalry, a new frame
The two billionaires fell out in 2015, after Musk recruited a top Google AI researcher to co-found OpenAI, according to Yahoo! Finance. So this is an unusual moment: a former rival's long-range forecast, praised publicly by the man who once pulled talent away from him.
It's tempting to read the praise as a peace signal. I'd be cautious. Talent fights in AI have been bitter for everyone involved, and a compliment on a forecast costs very little to give. The person making the compliment rarely pays a price if the forecast turns out wrong.
Here's the thing, though. The "five nines" figure is a secondhand account of a podcast answer, and the reporting attaches no timeline to the word "eventually." Treat it as a directional view of where AI compute might go, not a measurable target. The more useful test is whether orbital demos from Google and SpaceX turn into commercial capacity. That's the signal to watch, not the endorsements.
What this could mean for you
If you're working in this space, the useful question isn't whether orbital compute is coming. It's which workloads would move first. Latency-sensitive jobs probably stay on the ground, since the delay involved in reaching orbit adds up quickly. Batch processing, or inference that doesn't need to sit close to end users, is a more plausible early candidate.
Launch costs, radiation tolerance, hardware replacement, and getting data back down all sit between a demo and a working data center. Ambitious infrastructure bets have a habit of taking longer than their most optimistic backers suggest. That's not a knock on the idea. It's a reason to read timelines with some suspicion.
My take: this matters more as a signal about where two rival firms are placing bets than as a prediction about the future. Expensive. Unproven at scale. Still worth watching.
If you want to judge progress, skip the podcast clips and look at the boring metrics. Cost per unit of compute, uptime, and whether anyone signs a paying customer will tell you far more than the next endorsement will.
