Rob Carlson
@robcarlson.bsky.social
I aspire to leave the planet in better shape than I found it.
Managing Director, Planetary Technologies: @planetaryops.bsky.social
2024 Bloomberg Green Champion
Accidental Affiliate Prof of Computer Science, @UWCSE.bsky.social
Home: synthesis.cc
Managing Director, Planetary Technologies: @planetaryops.bsky.social
2024 Bloomberg Green Champion
Accidental Affiliate Prof of Computer Science, @UWCSE.bsky.social
Home: synthesis.cc
Reposted by Rob Carlson
Here he is predicting "SpaceX will launch and operate more AI compute annually than the cumulative total on Earth [requiring] 10K Starship launches per year... possibly 30K."
10K/yr is ~30/day; >1/hr. Even with 2 launch sites, this seems logistically optimistic.
science.slashdot.org/story/26/02/...
10K/yr is ~30/day; >1/hr. Even with 2 launch sites, this seems logistically optimistic.
science.slashdot.org/story/26/02/...
Musk Predicts SpaceX Will Launch More AI Compute Per Year Than the Cumulative Total on Earth - Slashdot
Elon Musk told podcast host Dwarkesh Patel and Stripe co-founder John Collison that space will become the most economically compelling location for AI data centers in less than 36 months, a prediction...
science.slashdot.org
February 5, 2026 at 8:54 PM
Here he is predicting "SpaceX will launch and operate more AI compute annually than the cumulative total on Earth [requiring] 10K Starship launches per year... possibly 30K."
10K/yr is ~30/day; >1/hr. Even with 2 launch sites, this seems logistically optimistic.
science.slashdot.org/story/26/02/...
10K/yr is ~30/day; >1/hr. Even with 2 launch sites, this seems logistically optimistic.
science.slashdot.org/story/26/02/...
Here he is predicting "SpaceX will launch and operate more AI compute annually than the cumulative total on Earth [requiring] 10K Starship launches per year... possibly 30K."
10K/yr is ~30/day; >1/hr. Even with 2 launch sites, this seems logistically optimistic.
science.slashdot.org/story/26/02/...
10K/yr is ~30/day; >1/hr. Even with 2 launch sites, this seems logistically optimistic.
science.slashdot.org/story/26/02/...
Musk Predicts SpaceX Will Launch More AI Compute Per Year Than the Cumulative Total on Earth - Slashdot
Elon Musk told podcast host Dwarkesh Patel and Stripe co-founder John Collison that space will become the most economically compelling location for AI data centers in less than 36 months, a prediction...
science.slashdot.org
February 5, 2026 at 8:54 PM
Here he is predicting "SpaceX will launch and operate more AI compute annually than the cumulative total on Earth [requiring] 10K Starship launches per year... possibly 30K."
10K/yr is ~30/day; >1/hr. Even with 2 launch sites, this seems logistically optimistic.
science.slashdot.org/story/26/02/...
10K/yr is ~30/day; >1/hr. Even with 2 launch sites, this seems logistically optimistic.
science.slashdot.org/story/26/02/...
Reposted by Rob Carlson
More thoughtful reporting on data centers in space.
But there is a physics problem that all these companies have to solve (next posts).
It would make way more sense to build data centers on Luna b/c then you can use the moon as a heat sink.
(H/t @eaterofsun.bsky.social)
bsky.app/profile/lore...
But there is a physics problem that all these companies have to solve (next posts).
It would make way more sense to build data centers on Luna b/c then you can use the moon as a heat sink.
(H/t @eaterofsun.bsky.social)
bsky.app/profile/lore...
Did a dive into the engineering challenges posed by data centers in space. They could work! But there's a lot of technological development to be done www.bloomberg.com/news/article...
Why AI Data Centers in Space Aren’t Totally Science Fiction
Tech billionaires want to move AI’s massive energy demands off Earth. The engineering hurdles are enormous.
www.bloomberg.com
January 13, 2026 at 6:16 PM
More thoughtful reporting on data centers in space.
But there is a physics problem that all these companies have to solve (next posts).
It would make way more sense to build data centers on Luna b/c then you can use the moon as a heat sink.
(H/t @eaterofsun.bsky.social)
bsky.app/profile/lore...
But there is a physics problem that all these companies have to solve (next posts).
It would make way more sense to build data centers on Luna b/c then you can use the moon as a heat sink.
(H/t @eaterofsun.bsky.social)
bsky.app/profile/lore...
Reposted by Rob Carlson
Or you could go the other way, and for a fixed size of radiator on orbit you could improve processor efficiency by plugging in new 20ft TEUs (based on image above).
But now your thermal budget limits your competitiveness.
Again, seems better to go straight to Luna (or another large heat sink).
But now your thermal budget limits your competitiveness.
Again, seems better to go straight to Luna (or another large heat sink).
January 13, 2026 at 6:16 PM
Or you could go the other way, and for a fixed size of radiator on orbit you could improve processor efficiency by plugging in new 20ft TEUs (based on image above).
But now your thermal budget limits your competitiveness.
Again, seems better to go straight to Luna (or another large heat sink).
But now your thermal budget limits your competitiveness.
Again, seems better to go straight to Luna (or another large heat sink).
Reposted by Rob Carlson
The size of a data center in earth orbit with N processors scales roughly as N^(1/3), but the size (and thus mass) of the radiator scales as N (maybe faster).
As you double N, the radiator becomes an increasingly difficult (i.e. expensive) launch and operational problem compared to the processors.
As you double N, the radiator becomes an increasingly difficult (i.e. expensive) launch and operational problem compared to the processors.
January 13, 2026 at 6:16 PM
The size of a data center in earth orbit with N processors scales roughly as N^(1/3), but the size (and thus mass) of the radiator scales as N (maybe faster).
As you double N, the radiator becomes an increasingly difficult (i.e. expensive) launch and operational problem compared to the processors.
As you double N, the radiator becomes an increasingly difficult (i.e. expensive) launch and operational problem compared to the processors.
The size of a data center in earth orbit with N processors scales roughly as N^(1/3), but the size (and thus mass) of the radiator scales as N (maybe faster).
As you double N, the radiator becomes an increasingly difficult (i.e. expensive) launch and operational problem compared to the processors.
As you double N, the radiator becomes an increasingly difficult (i.e. expensive) launch and operational problem compared to the processors.
February 4, 2026 at 5:03 PM
Reposted by Rob Carlson
So, #energysky peeps: is there a CAISO version of this plot? CAISO must have equivalent (or better) data. Presumably it would show that 1) wholesale prices in CA peak in the shoulder hours when they burn gas and import electricity, and 2) batteries are lowering prices?
@mzjacobson.bsky.social?
@mzjacobson.bsky.social?
February 3, 2026 at 11:51 PM
So, #energysky peeps: is there a CAISO version of this plot? CAISO must have equivalent (or better) data. Presumably it would show that 1) wholesale prices in CA peak in the shoulder hours when they burn gas and import electricity, and 2) batteries are lowering prices?
@mzjacobson.bsky.social?
@mzjacobson.bsky.social?
So, #energysky peeps: is there a CAISO version of this plot? CAISO must have equivalent (or better) data. Presumably it would show that 1) wholesale prices in CA peak in the shoulder hours when they burn gas and import electricity, and 2) batteries are lowering prices?
@mzjacobson.bsky.social?
@mzjacobson.bsky.social?
February 3, 2026 at 11:51 PM
So, #energysky peeps: is there a CAISO version of this plot? CAISO must have equivalent (or better) data. Presumably it would show that 1) wholesale prices in CA peak in the shoulder hours when they burn gas and import electricity, and 2) batteries are lowering prices?
@mzjacobson.bsky.social?
@mzjacobson.bsky.social?