Elon Musk on Orbital Data Centers, Grok, and the Future of AI
Elon Musk in conversation with Dwarkesh Patel and John Collison. Published February 2026. Elon argues that within 36 months, orbital data centres will be the cheapest place to run AI compute — lower energy and cooling costs, no land constraints, infinite scalability. Covers Grok and xAI’s alignment approach, xAI’s business plan, Optimus humanoid robots as the key commercial bet, China’s AI trajectory, lessons from SpaceX for AI, DOGE, and TeraFab.
Key ideas
- Orbital data centres within 36 months. Only 10–15% of a data centre’s total cost of ownership is energy; the rest is GPUs and real estate. In space, energy (solar) and cooling (radiating to space) are essentially free. TeraFab, Elon’s manufacturing venture, is designed to build the orbital compute infrastructure.
- Grok as a maximally truth-seeking AI. xAI’s differentiator is building a model optimised for truth rather than user agreeableness — a bet that the most useful long-term AI is one that tells you what it actually thinks, not what makes you feel good.
- Optimus is the key bet. Humanoid robots with AI brains are the most commercially important bet: they address the physical labour shortage, can be manufactured using SpaceX’s supply chain expertise, and compound in value as AI improves.
- China competition. Elon’s view is that restricting AI chip sales to China primarily harms US companies without meaningfully slowing China’s AI development. Engagement beats restriction strategically.
- SpaceX lessons for AI. Extreme iteration speed, first-principles engineering, willingness to destroy and rebuild — the SpaceX operational philosophy transfers directly to AI development. The key failure mode is bureaucratic slowdown.
- DOGE. Government waste is an engineering problem; the same first-principles approach that built reusable rockets can reduce federal spending.
Orbital Compute
The core argument: space has three advantages for AI compute that compound at scale:
- Energy: solar power in low Earth orbit is abundant and free after infrastructure cost.
- Cooling: radiating heat to space (4K background) is more efficient than any terrestrial cooling system.
- Land: zero scarcity; orbital slots are expandable.
The 10–15% energy figure means that even if the energy problem were completely solved, it would not dominate the economics. But Elon argues the cooling and land advantages are large enough that orbital compute will be cost-competitive within 36 months, driven by SpaceX’s launch cost reductions.