Linklog 7

There’s so much narrative happening around AI.

I don’t know that SpaceX can afford the capex to compete with hyperscalers on raw capacity or cost anytime soon – at least, not while they’re still getting Starship ready, building out Starlink, and trying to get their orbital data centers going – but they’re certainly willing to compete on speed and throw terrestrial fossil-powered compute up pretty quick. Anyway, some armchair vibe-predictions:

1. A majority of SpaceX’s revenue next year will come from terrestrial data center contracts. This seems locked in by their deals with Anthropic and Google and realistic Starlink growth.
2. SpaceX will not release an open model next year.

Let’s see what happens!

TMF Associates blog » SpaceX’s Rorschach test. “It’s also undeniable that SpaceX needs more than just Starlink to justify a $1.5T valuation, given that even its expected lead investment bank, Morgan Stanley, only thinks Starlink revenues will get to $126B in 2040.” – This is the sharpest thing I’ve read about the SpaceX IPO yet. OP

Ways to think about token pricing — Benedict Evans. Also unclear to me: in whatever long-run future we get to, will uncertainty over when AI lab revenues land remain large relative to what they can hedge? This seems to determine the premia the big labs will pay for compute and the kinds of power infrastructure we’ll have in, say, a decade. OP

The future of work is world models. I don’t know if I agree with this yet, but it’s easy to see the appeal. OP