British astronaut Tim Peake envisions an period the place data centers in house grow to be a actuality, and he’s a person on a mission as he partnered with Axiom House to make orbital AI farms a viable possibility.
(Now we have lined Axiom House a number of instances on TechRadar Professional. Once they launched a prototype with Red Hat back in March, crossed path with AWS at Re:Invent in 2022 and once they teamed up with Nokia to add 4G LTE capabilities to spacesuits.)
I met him final Friday on the Goodwood Competition of Pace, the place he has been an envoy for FOS’s Future Lab, an immersive STEM-focused exhibition overlaying robotics, mobility, and house, for the previous eight years.
Past his involvement with Axiom House, I used to be notably eager to listen to his ideas about AI. “I exploit ChatGPT in all probability as a lot as different individuals do”, he advised me, “What I discover is that if I am asking it a query on one thing I do know fairly a bit about, however I would similar to to see what ChatGPT does, it is fascinating to see the construction and the logic and the thought processes concerned.”
He’s nevertheless, optimistic. “It’s maybe essentially the most highly effective instrument that we have ever developed, who is aware of?”, he candidly added, “And perhaps it’s kind of like having a calculator or a watch that does 1,000 features. In the mean time, we’re taking part in with the primary three or 4 of these features, probably not understanding the opposite 997 features that it might do. We’ve not but opened the instruction guide and actually dug into it that a lot”.
However it’s removed from excellent and Tim Peake got here up with this nice instance for instance the form of errors ChatGPT – and certainly every other AI instrument – could make. “You ask ChatGPT right this moment what instrument did Tim Peake play on the Worldwide House Station? It’ll come again and inform you I performed a saxophone. I did not play a saxophone in house. You understand, I performed a guitar. [French ESA astronaut] Thomas Pesquet performed a saxophone. I do know that as a human as a result of I spoke to Thomas.”
Bytes in house
The reliability – and high quality – of responses produced by generative AI was not the one concern for the Future Lab ambassador. Information facilities, he stated, are “heating huge quantities of water, as we use the water for cooling. We’re merely transferring that power into our water, which must reject that warmth someway. So we’re once more contributing to local weather change issues. And it is costing some huge cash.”
Information facilities in low earth orbit (LEO) make sense: thermal rejection in house is much less of an issue than on Earth (however nonetheless an issue as there’s no conduction or convection occurring) and the Solar can present limitless, low cost, clear power provide all 12 months spherical (however you’ll nonetheless want photo voltaic panels).
The sport changer is the commoditization of house journey, made attainable by personal corporations comparable to House X and Blue Origin.
They’ve slashed the price of sending moderately heavy payloads in orbit. A lot in order that House X, for instance, has a web based calculator the place anybody can transparently ebook flights, such as you’d do to your vacation. On the time of writing, sending 100Kg in LEO by November 2025, costs $650,000.
He revealed that Axiom House plans to launch two orbital knowledge middle nodes later this 12 months and expects orbital knowledge facilities to succeed in value parity with their earthly brethren inside a decade. Axiom House just isn’t the one one capturing for the celebrities.
One other US-based firm. Lonestar, is exploring the concept of placing a data center on the moon no much less. It suffered a big setback in March 2025 when its Freedom Information Middle payload didn’t land after Athena, the spacecraft carrying it, toppled over. Lonestar’s web site hasn’t been up to date since however the firm goals in the long term to have a “collection of ever more and more succesful multi-petabyte knowledge storage spacecraft [sic]”.
Let’s discuss energy
There’s one thing else, greater than house journey and AI farms that Tim Peake is completely obsessed with. If he had been to take a position £1 million in an rising expertise, it will be in fusion, as in nuclear fusion. “I believe it is our best problem.” he stated, “I believe it will likely be game-changing to stay in a future the place we do not even query the truth that everyone has low cost, limitless, clear power. Think about that. And so they’ll look again to our time and pondering, “Yeah, again within the days once we had been burning fossil fuels.” And that, you realize, this can be a small piece of historical past which is about to go extinct. And excitingly, I believe it will go extinct pretty quickly.”.
Don’t get me fallacious, he added, “house journey is thrilling. There’s quite a bit going to occur. And it is type of an incremental course of. However I do not assume there’s nothing revolutionary within the medium time period. Whereas with power, within the medium time period, I believe we’ll be revolutionary.”
Within the close to future, a not-so-insignificant portion of the world’s electrical energy manufacturing (up to 14% of the total US output according to a 2024 report from research firm Semianalysis) will feed hyperscalers and AI. And that’s one thing that puzzles Tim, the environmentalist.
“I can not fairly see the enterprise mannequin as to who’s working these programs. The truth that they (OpenAI) are fairly pleased with, you realize, billions and billions of individuals world wide anytime, in a position to demand a lot power from the programs that they are offering without spending a dime.”
And it’ll worsen as soon as Agentic AI turns into mainstream. “The tempo of studying and the amount of knowledge you could analyze goes to extend exponentially. However on the expense of Earth’s sources.”
PS: The complete 30-minute recording of my dialog with Tim Peake was transcribed by Google AI Studio and formatted by Google Gemini.
Source link
