MOUNTAIN VIEW, Calif., February 9, 2026 — Elon Musk believes the relentless race to build bigger AI data centers is fundamentally misguided. His solution? Take the show off-world.
AI’s Power Problem: Why Musk Is Looking to Space
Musk argues that current power constraints make scaling AI on Earth unsustainable, and proposes orbiting data centers powered by space-based solar energy.
- Musk estimates AI scaling will require doubling the entire US’s current power consumption.
- He proposes space-based solar power as a 10x cheaper alternative to terrestrial solutions.
- Musk predicts space will be the most economically viable location for AI infrastructure within 30-36 months.
- He is building a vertically integrated system—SpaceX, Tesla, and xAI—to overcome Earth’s physical limitations.
During a three-hour interview on John Collison’s Cheeky Pint podcast, co-hosted by Dwarkesh Patel, Musk detailed the rationale behind merging SpaceX with his artificial intelligence company, xAI. The core issue, he explained, is a looming energy crisis. Scaling AI to its full potential will require an unprecedented surge in electricity, potentially doubling the entire US’s current power usage of roughly half a terawatt.
The problem isn’t just demand. Chip production is accelerating exponentially, but power generation outside of China remains stagnant. While natural gas is an ideal fuel source for AI training, gas turbine manufacturers are currently backlogged until 2030. “Those who have lived in software land,” Musk stated, “don’t realize they’re about to have a hard lesson in hardware.”
Musk’s proposed solution is, characteristically, ambitious: move data centers into orbit. He argues that solar panels in space generate approximately five times more power than those on Earth, unhindered by atmospheric interference, clouds, or nighttime. His estimate is that space-based solar power would be ten times cheaper than building out terrestrial infrastructure.
“You can mark my words,” Musk declared. “In 36 months, probably closer to 30 months, the most economically compelling place to put AI will be space.”
Musk acknowledged the challenges inherent in such a rapid timeline, noting that his projections as an executive are intentionally aggressive to push for speed. He admitted that many factors could derail the plan, but believes it remains achievable.
Beyond Power: Chips, Robots, and China
The interview also covered several other key areas of Musk’s ventures. He revealed plans for “TeraFab,” a chip factory designed to produce over 100 million advanced chips annually, recognizing that chip availability will become a critical bottleneck within five years if power constraints are resolved.
Developing the Optimus humanoid robot has also presented significant hurdles, requiring the creation of custom actuators, motors, gears, and sensors due to the lack of an existing supply chain.
Musk offered a stark assessment of the geopolitical landscape, stating, “We can’t win on the human front, but we might have a shot on the robot front.” He pointed out that China currently produces three times the electricity of the US and controls 98% of global gallium refining—a crucial element in high-end electronics.
The Bigger Picture: A Vertically Integrated AI Strategy
Musk’s companies are now strategically aligned to address the challenges of AI scaling. SpaceX will launch the necessary satellites, Tesla will manufacture the solar panels and robots, and xAI will develop the artificial intelligence itself. This vertically integrated system is designed to overcome Earth’s physical limitations and maximize solar energy capture, potentially propelling humanity up the Kardashev scale.
Musk predicts that SpaceX will launch more AI compute annually than the combined total on Earth within five years. Whether or not this timeline proves accurate, the fundamental constraint remains: chips are useless without power, and we are rapidly approaching the limits of terrestrial power infrastructure. If space-based solar power can provide a solution, the primary limitation will shift to chip production.
A full transcript of the interview is available on Dwarkesh’s blog.
