108
AIA fridge in space just trained an AI. No, really. 🛰️
A fridge-sized satellite with an Nvidia H100 GPU launched into orbit in November 2024. It trained a small language model on Shakespeare. And the engineers were surprised, not because it worked, but because it worked better than expected.
Meanwhile, back on Earth: 30–50% of all planned data centre capacity for 2026 is delayed. Permits. Grid connections. Power shortages. Local opposition. The physical infrastructure underpinning the AI boom is quietly running into a wall.
Here's where it gets interesting for business leaders 👇
The cost gap between orbital and terrestrial data centres is closing faster than anyone predicted. With optimised satellite design, we're already within 5% parity. If Starship brings launch costs down to $200/kg (from $1,500 today), orbital data centres become cheaper than building on Earth — $12.1B vs $15.9B for 1GW of capacity.
This isn't science fiction. It's infrastructure arbitrage. When land, power, and planning permission become the scarce resource, orbit becomes the workaround.
The businesses that understand this now — and are watching Starship's test flight schedule as closely as their cloud billing — will have a structural advantage in 2027 and beyond.
Full analysis on LinkedIn → https://www.linkedin.com/feed/update/urn:li:activity:7434680491267604480
Weekly AI business insights → https://aidrivesbiz.com/
What's your read on this, overhyped moonshot, or genuine infrastructure play? 👇 Tag a founder or CTO who should be watching Starship's next launch for business reasons, not just space ones.
Follow @AIdrivesBiz for daily AI business trends
@aidrivesbiz
