San Francisco: Tesla and SpaceX CEO Elon Musk has predicted that space could become the cheapest location to run artificial intelligence infrastructure within the next 36 months, arguing that orbital data centres would outperform Earth-based facilities in both energy efficiency and scalability.
Speaking on the Dwarkesh Podcast, Musk discussed the long-term economics of AI compute, power generation challenges on Earth, GPU reliability, and the possibility of building large-scale data centres in orbit. He claimed that solar power generation in space could be significantly more efficient and remove several constraints that currently affect terrestrial AI facilities.
“It’s harder to scale on the ground than it is to scale in space,” Musk said, outlining a future where AI training clusters and compute infrastructure may be deployed beyond Earth’s atmosphere.
Solar efficiency in space much higher, says Musk
Based tech billionaire Musk said one of the biggest advantages of space-based AI infrastructure would be energy generation through solar panels. According to him, solar arrays in orbit could deliver far greater output compared to ground installations.
“You’re also going to get about five times the effectiveness of solar panels in space versus the ground, and you don’t need batteries,” he said.
He explained that several Earth-side limitations reduce solar efficiency, including:
- Day–night cycles
- Seasonal variation
- Cloud cover
- Atmospheric filtering
“The atmosphere alone results in about a 30 per cent loss of energy,” Musk noted. Without atmospheric interference and with near-constant sunlight exposure in certain orbital paths, solar panels could operate at much higher and more predictable output levels.
He added that eliminating dependence on large battery storage systems — which are currently necessary for night-time and low-generation periods on Earth — would further reduce infrastructure costs.
Claim: AI compute costs may drop in orbit
Entrepreneur Musk argued that once launch costs and orbital construction become more efficient, placing AI compute clusters in space could become economically attractive.
“So any given solar panel can do about five times more power in space than on the ground. You also avoid the cost of having batteries to carry you through the night. It’s actually much cheaper to do in space,” he said.
Based on current technology trends and launch capabilities, Musk predicted a relatively near timeline for such a shift.
“Space will be by far the cheapest place to put AI. It will be space in 36 months or less — maybe 30 months,” he said.
While he did not provide a detailed deployment roadmap, Musk’s companies — particularly SpaceX with its heavy-lift rockets and Starship programme — are among the few private players with the launch capacity that could theoretically support orbital industrial infrastructure.
Scaling limits on Earth a growing concern
Tech leaders, including Musk, have increasingly highlighted the power and cooling demands of AI data centres as a major bottleneck. Advanced AI model training requires massive GPU clusters that consume large amounts of electricity and generate substantial heat.
Globally, AI data centre expansion has begun to strain local power grids, water supplies (for cooling), and land availability. Industry analysts have warned that energy access — not chips — could become the main limiting factor in AI growth over the next decade.
Musk suggested that space-based systems could bypass many of these constraints, offering effectively unlimited solar energy and natural radiative cooling.
Musk downplays large-scale GPU failure risks
Technology discussions around AI infrastructure often focus on hardware reliability at scale. Addressing this, Musk said GPU failure rates during large AI training runs are often overstated.
When asked about reliability concerns, he said performance depends partly on how new the hardware is when deployed.
“It depends on how recent the GPUs are that have arrived,” Musk said. He noted that most instability occurs early in the lifecycle, during initial deployment and debugging.
“Once they start working and you’re past the initial debug cycle — whether it’s Nvidia, Tesla, or other chips like TPUs or Trainium — they’re quite reliable past a certain point,” he said.
Because of this, Musk suggested that ongoing servicing requirements for large AI clusters may be more manageable than critics assume — a factor that would matter even more for space-based systems where physical repair is difficult and expensive.
Manufacturing and robotics link to AI expansion
Podcast discussions also touched on Musk’s broader vision of scaling advanced manufacturing and humanoid robots in the United States. He linked AI growth with robotics and automated manufacturing, suggesting future factories could produce both compute systems and robots at very high volume.
Such manufacturing capacity, he implied, would be necessary to support rapid AI infrastructure expansion — whether on Earth or in orbit.
Conclusion
Musk’s prediction that space could soon become the cheapest location for AI infrastructure reflects growing concern over Earth-based energy and scaling limits. While significant engineering, cost, and regulatory hurdles remain before orbital data centres become practical, the idea signals how aggressively AI leaders are rethinking infrastructure. If launch costs continue to fall and space-based power systems mature, off-planet computing may shift from science fiction to pilot projects within the next few years.