Orbital Data Center cost takes by Street analysts making the rounds:
Wood Mac ~$170B/GW,
BofA $160B to 180B,
Futurum ~$72B (infrastructure alone),
BNP >$100B just for launch to loft 1 GW.
This is all the same spreadsheet genre that “proved” Starlink couldn’t work in comparison to classical Data Centers at $40B-$60 on the ground.
However, we now know that Starlink worked because:
(a) SpaceX’s real Falcon 9 cost was nowhere near list price, and
(b) V2/V3 satellites blew past the bandwidth per satellite that those early models assumed.
Starlink satellites kept outrunning the forecast and internal launch costs ran way below analysts estimates.
AI satellites will rhyme.
Gen-1 is what the cost models price.
Gen-2 and Gen-3 (better chips, power, and thermal per kg) are where we once again pull away from the spreadsheet.
Starship does the launch cost side slide in short order. @Mach33 is the most comprehensive analysis on Data Centers in the sky. Here are their estimates:
Starship AI launches, with booster reuse and ship expendable cost at ~$300/kg.
That would give us:
~$46B per GW for SpaceX, then
toward ~$24B/GW over the next decade with Ship reuse.
Bull case with full reuse and better chips goes lower.
In summary: actual cost likely to start at $46B per GW (and the falling to half that) vs $160B+ headlines from "analysts".
Add to that less permitting delays and less political pushback than a GW on the ground.
Starlink redux.
Bears beware.