Elon Musk is known for bold predictions, but his latest warning about artificial intelligence may be one of his most dramatic. He argues humanity is rapidly approaching a hard physical limit—not software or algorithms, but electricity.

Speaking on the Cheeky Pint podcast with Dwarkesh Patel, Musk said the explosive growth of AI could overwhelm Earth’s power capacity within months, not decades. In his view, energy—not intelligence—is the real bottleneck.

Musk noted that while debates focus on AI ethics and safety, massive data centers already consume extraordinary amounts of electricity. Scaling power generation fast enough, he argues, is far harder than scaling AI itself.

“All of the U.S. uses about half a terawatt of power,” Musk said, questioning how realistic it is to double that quickly. He believes Earth-based infrastructure may simply not keep up.

His proposed solution is radical: move large-scale AI infrastructure into space. Musk predicts that within 30 to 36 months, orbit will become the cheapest and most practical place to run advanced AI systems.

The reason, he says, is solar power. Solar panels in space receive near-constant sunlight, avoiding clouds, night cycles, and atmospheric loss. Without the need for batteries, orbital solar could be far more efficient.

Musk has reportedly filed plans that could eventually involve up to one million satellites supporting space-based, solar-powered data centers—an unprecedented scale compared to today’s roughly 15,000 orbiting satellites.

Critics point to risks such as space debris, regulation, and geopolitics, while others argue Earth-based solutions like nuclear power could still meet demand. Still, Musk’s warning highlights a real issue: AI’s future is constrained by physics, power, and planetary limits—and the next frontier may lie in orbit.