Elon Musk is quietly walking away from one of the founding ideas of his empire. According to TechCrunch AI, the SpaceX IPO filing released this week reframes solar power as a space business, not an Earth business, while xAI loads up on natural gas turbines to feed its data centers. For a CEO who built Tesla on the promise of moving the world off hydrocarbons, that’s a hard pivot worth taking seriously.
What stands out here is the gap between Musk’s stated climate mission and where his capital is actually flowing. TechCrunch AI reports that xAI is running dozens of unregulated gas turbines and has lined up $2.8 billion more in turbine spending. Meanwhile, xAI dropped $697 million on Tesla Megapacks and SpaceX bought 1,279 Cybertrucks for $131 million. Solar panels from Tesla? Barely a line item.
The new pitch: data centers in orbit
The SpaceX filing reframes terrestrial solar as a benchmark to beat. Space-based arrays, the company claims, can produce “more than five-times the energy” of ground installations thanks to round-the-clock sunlight. The logic is simple: AI compute demand is exploding, NIMBYs are blocking new sites, and the grid can’t keep up. So put the servers where the sun never sets.
Musk’s framing is aggressive. The filing references “terawatt-scale annual AI compute growth” as a planning assumption. Context: every data center on Earth today draws roughly 40 gigawatts combined. Musk is modeling a future where AI alone needs 25x that capacity added every year.
Why this matters now
Three forces are converging, and Musk is reading them as a single problem:
- Power scarcity: Hyperscalers are signing nuclear PPAs, reopening Three Mile Island, and burning gas because renewables can’t scale fast enough.
- Permitting drag: Local opposition to data centers is now a real bottleneck in Virginia, Ohio, and Texas.
- Compute demand: Frontier model training has no ceiling in sight, and inference is climbing alongside it.
Musk’s answer is to leapfrog the whole mess by moving compute off-planet. Other Silicon Valley executives, as TechCrunch AI notes, are kicking the same tires. Jeff Bezos has floated similar ideas. The orbital data center is becoming a respectable thesis, not a sci-fi punchline.
The skeptic’s case
TechCrunch AI’s analysis flags the obvious holes. Launching a panel to orbit burns more energy than trucking one to a solar farm in Nevada. Space-grade hardware costs multiples more than terrestrial gear. Starlink’s per-watt power costs already run far above what a Virginia data center pays. And it’s an open question whether AI training workloads can even be sharded across satellite constellations with the latency budgets they need.
The deeper critique: Musk is letting the perfect kill the good. Terrestrial solar is nowhere near tapped out. Battery storage is getting cheaper every quarter. The Master Plan Part 3, published just three years ago, laid out a credible roadmap to decarbonize the existing grid. Now xAI’s gas turbines are the counter-example to Musk’s own thesis.
Takeaways for operators
If you’re building or buying AI infrastructure, three things to watch:
- Power is the new constraint, not chips. GPU allocation matters less than megawatt allocation in 2026 planning.
- The clean-energy halo is fading from AI. Expect more gas, more nuclear, less solar in the next 24 months of hyperscaler buildouts.
- Orbital compute is a 10-year bet, not a quarterly one. Don’t model it into near-term capacity plans, but don’t dismiss it from long-range scenarios either.
Musk’s track record on spotting inflection points is real. So is his track record on timelines. The question isn’t whether space-based solar eventually happens. It’s whether xAI’s gas turbines are a two-year bridge or a two-decade lock-in. Full details at TechCrunch AI.