Opinion piece. The views below are mine and concern a technology, not any security.
On November 6, 2026, Elon Musk replied with a single word, “Yes”, to a post on X that replays one of his favorite predictions: within two to three years, space will be the cheapest place on Earth (well, off Earth) to run artificial intelligence.
Elon Musk predicts space will become the lowest-cost place to run superintelligence within just 2–3 years
— X Freeze (@XFreeze) November 5, 2026
“I think the case is a no-brainer for building solar-powered AI data centers in space”
Solar panels facing the Sun and radiators pointed away from it can provide an incredibly efficient power and cooling system
“The net effect is that the lowest cost place to put AI will be space. And that’ll be true within two years, maybe three at the latest”
https://t.co/SkjFq4Qg8z
Yes
— Elon Musk (@elonmusk) November 6, 2026
One detail matters here. The quote in that post is not new. Musk said it in January 2026 at the World Economic Forum in Davos, on stage with BlackRock’s Larry Fink, as pv magazine reported at the time. So the “two years, maybe three” clock did not start this week. It started in January. If he is right, orbit should beat the ground on cost somewhere between early 2028 and early 2029.
Ten months in, I wanted to check where that clock actually stands.
The case Musk makes
The argument is simple, and honestly, it is elegant. A solar panel in orbit gets sunlight without clouds, weather or night if you pick the right orbit. Musk puts the gain at about five times the energy of the same panel on the ground. Point the panels at the Sun, point the radiators at deep space, and you get power and cooling without a grid connection, without water and without a local planning committee.
He has repeated the idea several times since Davos. When SpaceX absorbed xAI in February, he wrote on the SpaceX website that the lowest-cost way to generate AI compute would be in space within two to three years, according to the Associated Press. On Dwarkesh Patel’s podcast the same month, he gave himself 30 to 36 months.
Where the clock actually stands
Here is what has really happened on the hardware side, as far as I can verify:
- Starcloud, a venture-backed startup, launched a satellite carrying an Nvidia H100 chip in November 2025 (NPR).
- Google put a refrigerator-sized test satellite with AI chips into orbit for Project Suncatcher, NPR reported on October 1, 2026.
- SpaceX has said it expects to start deploying “orbital AI compute satellites” as early as 2028 (NPR).
So we have prototypes. Real ones, which is more than a slide deck. But a prototype with a handful of chips and a data center that competes on price with a gigawatt campus in Texas are two very different animals.
Why I am not convinced yet
The most interesting skeptic is not a rival. It is SpaceX’s own IPO prospectus. In its S-1 registration statement, filed with the SEC on May 20, 2026, SpaceX says its orbital AI compute plans, along with its Moon and Mars ambitions, involve “unproven technologies or technologies that do not exist”. It also warns that it may not be possible to put a date on 100 gigawatts of compute in orbit. Lawyers write risk factors to protect the company, so I would not read that as a verdict. But it is a long way from “no-brainer”.
Then there are the numbers that keep me up at night (not really, I sleep fine):
- Energy is only part of the bill. Dwarkesh Patel pointed out on his podcast that energy is roughly 10 to 15% of a data center’s total cost of ownership. The chips are most of the rest, and in orbit they cannot be repaired. The S-1 says SpaceX does not plan to service or repair processors in space.
- Launches. Patel estimated that 100 gigawatts in orbit means on the order of 10,000 Starship launches. Morningstar’s estimate, cited by R&D World, is about 6,667 flights a year, far beyond anything planned for the near term, in its view.
- Heat. In a vacuum, there is no air to carry heat away. Everything leaves by radiation, which is why the S-1 says compute satellites would need much bigger radiators than today’s Starlink satellites. For scale, R&D World points out that the International Space Station dumps tens of kilowatts of heat, while an AI cluster would have to dump megawatts.
- Not everyone agrees. Microsoft president Brad Smith said he would be surprised if people moved from land to low-Earth orbit (Associated Press).
My take
I think the physics argument is real. Free sunlight and free cold are not science fiction, and the fact that Google, SpaceX and several startups are all putting hardware in orbit tells me this is more than a meme.
But “cheapest place to run AI by 2028 or 2029” is a different claim from “technically possible”. To get there, Starship has to fly fully reusable at a cadence nobody has ever seen, chips have to survive years of radiation with zero repairs, and radiators have to scale by orders of magnitude. Each of those is a big bet. Musk needs all three at once.
Musk also has a track record of being right on the destination and optimistic on the date. So my personal reading is this: the “Yes” is a reaffirmation, not news. The real news will be the first orbital cluster that sells compute at a price a customer can compare with the ground. When that happens, I will write about it. Until then, I am keeping the January 2026 date in my calendar and checking back.
What do you think: orbital data centers by 2029, or a 2035 story?
Sources
- pv magazine, Musk says solar space-based AI data centres possible in two to three years (January 28, 2026)
- Associated Press via The Journal, Musk joins his rocket and AI businesses into a single company before an expected IPO this year (February 2026)
- SpaceX (Space Exploration Technologies Corp.), Form S-1 registration statement, SEC EDGAR (May 20, 2026)
- R&D World, Musk says the cheapest place to put AI is space. SpaceX’s IPO filing isn’t so sure. (2026)
- NPR, Google launches Project Suncatcher, a step towards AI data centers in space (October 1, 2026)
- Posts on X by @XFreeze (November 5, 2026) and @elonmusk (November 6, 2026), embedded above
AI disclosure: This article was drafted with AI tools (Claude, by Anthropic) from the sources listed above and reviewed by the editor before publication.
Disclaimer: This article is for information and entertainment purposes only. It is not investment advice and not a recommendation to buy, sell or hold any security. Past performance does not guarantee future results, and investing involves a risk of loss. Disclosure: At the time of writing, the author holds shares of Alphabet and Microsoft. TickerRoast receives no compensation from the companies discussed.
Read next
Get the daily market briefing
The stories that move markets, in your inbox every trading day at 9 a.m. ET. Free, unsubscribe anytime.
