In partnership with
Three tech giants are buying nuclear reactors. Microsoft wants to fire up Three Mile Island again. Amazon paid $650 million for a data center sitting next to a Pennsylvania nuclear plant. Meta's in the market for its own.
When companies start shopping for decommissioned power plants, you know something broke in the planning process.
On January 30, SpaceX submitted an FCC filing for one million satellites. Not for internet. For processing AI workloads in orbit using pure solar power and space as a natural heat sink.
There are roughly 15,000 satellites circling the planet today. This would multiply that by 67.
The Infrastructure Bottleneck Nobody Planned For
Training GPT-4 consumed enough electricity to power 1,000 homes for a year. GPT-5 is even more worse. Data centers already eat 2% of global electricity, and that number grows every month.
Chips aren't the problem. TSMC is making plenty of GPUs. The issue is electricity and heat. Power companies are literally saying no to hyperscalers asking for more capacity. It'll be years before they can add more. Meanwhile, cooling systems drain water supplies that most places can't spare.
The bottleneck isn't money or talent anymore. It's watts and cooling. Space solves both. Unlimited solar power, natural cooling, zero grid competition. The physics work. Whether the economics do depends on launch costs.
SpaceX is solving compute infrastructure in space. Deel solves hiring infrastructure on Earth. Try Deel here ⤵
AI in HR? It’s happening now.
Deel's free 2026 trends report cuts through all the hype and lays out what HR teams can really expect in 2026. You’ll learn about the shifts happening now, the skill gaps you can't ignore, and resilience strategies that aren't just buzzwords. Plus you’ll get a practical toolkit that helps you implement it all without another costly and time-consuming transformation project.
Built on Proven Infrastructure
SpaceX operates 6,000 Starlink satellites providing broadband to over 100 countries. They've launched more satellites than the rest of the world combined over the past three years. The supply chain exists. The launch cadence exists. The operational experience exists.
The proposed orbital data centers use Starlink V3 satellites with 1 terabit per second throughput, ten times current capacity. These already communicate via optical inter-satellite laser links at 200 Gbps, creating a mesh network that routes data between satellites without ground station hops.
Elon Musk confirmed in November 2025 that SpaceX is scaling Starlink V3 for data processing applications.
Technical Foundation: Optical inter-satellite links operational since 2020. SpaceX manufactures satellites at scale and maintains the world's largest active constellation. The capability is proven, not theoretical.
Starship Economics Change the Calculation
Current launch costs make orbital data centers economically absurd. Getting payload to orbit runs thousands of dollars per kilogram. At that price point, terrestrial infrastructure wins despite power and cooling challenges.
SpaceX is making a different bet. Starship aims for full reusability with massive payload capacity. If successful, launch costs drop by an order of magnitude or more. At that price point, the equation flips.
Solar power in space becomes competitive with grid electricity when you factor in cooling costs and real estate. No permitting delays. No transmission losses. No dependence on regional power availability. The infrastructure becomes self-contained.
Privacy-first email. Built for real protection.
End-to-end encrypted, ad-free, and open-source. Proton Mail protects your inbox with zero data tracking.
Orbital Debris Concerns Are Real But Addressable
One million satellites raises an obvious question. What happens when they crash?
SpaceX included collision avoidance in the filing. Automated tracking, electric propulsion for dodging, shared orbital data with other operators. Dead satellites get parked instead of becoming debris.
But here's the thing. SpaceX asked for 42,000 Starlink satellites and deployed 6,000. Filing big is how negotiations work. Ask for a million, get approved for way less.
Debris is a concern. But launch economics and FCC approval matter more.
Negotiating Reality: One million is an opening position. The FCC will approve a fraction. But even 5% would be transformative for orbital computing capacity.
What This Signals About AI Infrastructure
The specific numbers matter less than what this reveals. AI labs are hitting power and cooling limits that money can't fix. Grid capacity takes years to build. Water is scarce in most places. Nuclear plants are expensive and politically toxic.
Space bypasses all of it. Companies are also exploring subsea data centers, geothermal facilities, and microreactors. But orbital computing is the most extreme option because it ignores Earth's constraints completely instead of working around them.
What's certain is that frontier labs need way more compute than current infrastructure can deliver. And they're willing to try things that would've sounded crazy three years ago.
Further Reading:
Creativity + Science = Ads that perform
Join award-winning strategist Babak Behrad and Neurons CEO Thomas Z. Ramsøy for a strategic, practical webinar on what actually drives high-impact advertising today. Learn how top campaigns capture attention, build memory, and create branding moments that stick. It’s all backed by neuroscience, and built for real-world creative teams.
Stay sharp,
Better Every Day
