ISSUE 013 · The Aroko Letter
No Exit to Orbit
A viral 1,000x claim runs up against the physical constraints that actually govern AI’s growth
A Cloudflare CFO floated a 1,000x growth number for non-human internet traffic; Musk repeated it with no analysis attached. This issue asks what 1,000x AI inference growth would actually require from the physical world — walking three plateaus over the next 25 years, each with its own binding chokepoint, through to the "leave the planet" argument for space-based compute, and back to why Northern Virginia keeps winning even as its own queues get worse.
Anchor figure
Key takeaways
- The physical ceiling on the AI buildout — how much power and material could theoretically support it — sits somewhere between 50 and 250 times today’s actual load, not close to 1,000x, and not yet close to that ceiling either.
- Three plateaus over 25 years: logistics (chips, transformers, queues) through 2030; geology and geopolitics (materials, export licensing) through 2040; physics itself (thermodynamic limits) through 2050.
- Northern Virginia’s data center vacancy is under 1% with 7-year interconnection waits — and its share of North American capacity keeps rising anyway. Scarcity clears through price and time, not exit.
- Even AGI arriving inside this window does not get to skip the physical stack: mined materials, built factories, and installed grid connections don’t move faster because the software asking for them got smarter.
Where it sits in the series
Names the Grid-Silicon Order directly as the physical-binding property at the center of the buildout — the throughline for every issue before it.
READ THE FULL ISSUE Published in full, free, on Beehiiv.
Forecasts this issue put on the watchlist
Open Power EquipmentThe Aroko Letter
Free. Biweekly. No sponsored research. Every forecast tracked in public with its kill condition attached.
GET THE LETTER