Experts Debate Space Data Center Feasibility On OpenAI Podcast

what drives me nuts on spacex coverage is all the talk of data centers in space but relatively little about how impractical that might be

latency for one is obvious. but this shit breaks all the time *on earth* and needs fixing

obvious valuation focus

I asked two of the foremost experts on data centers about putting them in space…

to be clear, as noted in replies, there are interesting technical challenges and theoretical use cases, which is why i think engineers are so enamored. hard problems are fun to attack

the context, however, is "i need to sell the public the largest valuation ever within a week"

@MikeIsaac It is stupid because all the technical problems are "solved" (cooling, power, hardening from micrometeoroids) but the costs are still like 5-10x earth and the only rocket which can launch them has a 58% launch rate.

@Deve_Dy @MikeIsaac the truly dumb explanation is the decision-makers have seen movies where astronauts out of airlock immediately freeze and think thats what happens

@MikeIsaac imagining a little piece of space junk makes a pinhole in the orbiting datacenter and all the air slowly leaks out. Pressure inside drops to zero, the vapor chambers of all the GPU heatsinks explode, whole thing melts down.

@valarauca1 i think normal people who dont really understand what is required to do this — and i BARELY understand but know enough to be dangerous — see the roadshow talk and think "well he did all this other stuff! he could do this maybe! what a dream!"

critical thinking out da window

@MikeIsaac The only thing really interesting about space datacenter is ~unlimited energy, but I'm curious how many square feet of solar panels you'd need to run a 1 megawatt nvidia rack (as the next generation is expected to be).

@MikeIsaac I know enough to ballpark the numbers, and nobody is lying, everyone is admitting how expensive it would be.

It is stupid. 5-10x the cost, 2-3x the risk it randomly explodes, 1.5-4x the latency, and 2-3x the lead time. There is no universe this makes a lick of economic sense.

@MikeIsaac FWIW, your nearest Starlink satellite is only ~1.6 milliseconds away (~3.2ms RTT based on ~480km LEO). Latencies to LEO are often better than latencies across terrestrial networks to one's nearest terrestrial data center.

@MikeIsaac I would like them to explain how they're going to cool these things when heat famously does not diffuse very much into vacuums

@tub_e_o "orbit" is an operative word here i.e. moving target

my main point is we're putting the cart so far ahead we cant even see the horse, and there's a reason, and that's because this stock needs to pop for them next week

@AndrewMayne so the experts said "probably not" without laughing in the interviewer's face

@MikeIsaac In addition to the question of HOW ARE YOU PEOPLE GONNA MAKE IT HAPPEN!!! Elmo also tried to dig tunnels and reinvent train tracks, dude is gonna fund one giant HDMI cable to space for this to work isn’t he…

@MikeIsaac Home solar panel is 67.8 in × 44.6 in for 450W. So assuming everything works as advertised you'd need a full acre of solar panels to run a single Vera Rubin Ultra rack?

@MikeIsaac *would be

@MikeIsaac It is probably just a cover to expand starlink, as you'd need more satellite <-> satellite bandwidth.

@Deve_Dy @MikeIsaac ive been taking the sunset limited for years to commute between new orleans and houston & after a day at the menil and museum of fine arts will sit down at west alabama icehouse for a cold one. it gets a big nasa crowd and theyre all 'this is a morons idea'

@AndrewMayne @JustinRYoung There's no water to poison in space. Executives won't go for it.

@MikeIsaac sycophants chasing a check & engineers having fun optimizing a problem without thinking of the realities.