Musk: Merging SpaceX and xAI, lol wut?

I woke up on Monday to the news that was merely a rumor last week, that Elon Musk on Friday officially merged two of his playthings, SpaceX, and xAI.

SpaceX is the largest private space launch company, it is privately held, and they are planning an IPO this summer (June? July?) to raise funds.

SpaceX is the reason why the government has to play nice with Elon, because he's the critical path to our launch to orbit.

Thanks to all the conservatives who have underfunded NASA and space exploration for decades. I digress.

Musk also was an original investor in OpenAI, but left in a snit because he couldn't have all the toys under his control. That is being litigated, and this spring there will be a jucily delish deposition of Musk, under oath, that should give us ample grist for blogging.

When he realized that ChatGPT was pretty cool, he huffed, and did what all rich bullies do and made his own chatbot, starting xAI. But he was late to the game and had to catch up, that meant poaching elite talent (very $$$$), and tossing beaucoup bucks at it.

Fun fact, last year, as the banks wanted their money back from the Twitter acquisition, and his co-investors didn't want to lose on their investments, he folded Twitter into xAI at par value (the $54B it was purchased for).

Because, why not.

xAI burns a bit over $1B a month. It drives the Grok chatbot, and the Grokipedia (a dollar store "unwokeified" Wikipedia) and it has been sold in large batches to the US Military.

Currently it was valued at ~$230B, but that is wall-street speak for "early growth stage startup" money.

So, naturally Musk felt that there were "synergies" to be had by merging SpaceX and xAI, and thus, prior to the IPO, why the fuck not.

Ostensibly, this is to build data centers in space, and that will make AI moar betterer.

If you haven't seen a data center in person, here's a typical one:

Inside, they are stuffed with racks and racks of computers and networking gear:

Indeed, the latest "rack" of Nvidia AI optimized compute and GPU cluster is about 3.5 tons (7,000 pounds, 3,200 kilograms). That is heavy as fuck. If you drop that on your toe, you can kiss that tootsie buh-bye.

Anyhow, these monstrosities need to be placed into a building, and it takes tens to hundreds of megawatts of power to run, and because the computers create waste heat, they use a lot of water to cool them. This impacts the communities where people live. They use a lot of electricity, so that raises the price per kWh normal people pay. And you know it by your utility bill. They use water – LOTS of water – in their heat exchangers to keep the silicon cool. That is drying up groundwater, drying up wells, and leaving reservoirs so low that silt and sludge gets delivered to residences.

So, a lot of communities are banding together to oppose and fight these monstrosities that seemingly exist to help students cheat, and to flood social media with AI slop content.

Musk has a brilliant idea. He will merge xAI and SpaceX to launch data centers into space.

Deloitte says:

By 2035, power demand from AI data centers in the United States could reach 123 gigawatts, up from 4 gigawatts in 2024.

And they are spending beaucoup bucks to make this happen.

So, how plausible is it that you can build a series of data centers in earth orbit? I am not an aerospace engineer, but I know a little bit about the tech, so I asked ChatGPT to do some targeted research on this, and here's the summary:

Bottom line:

A 1-GW orbital data center is not ruled out by physics, but it is a civilization-scale space industrial project relative to current orbital logistics.

Mass to orbit: ~25,000–45,000 tons

Launches (Starship @ 150 t): ~170–300 flights

Realistic delivered + assembled cost: ~$100B–$300B

Primary “could be impossible in 10 years” blocker: GW-scale space fission + multi-GW heat rejection reliability

"Civilization-scale" project, and for 1 gigawatt deployment would cost ~ $300B, and how many Starship's have actually made it to even low earth orbit to date? What, that's 0? Yep, that is the case.

If you want to read this thread, here's the chat, the more detailed cost and feasibility is the second prompt down, so read through.

Net/net: Musk ain't going to put any real data centers into space, but this was merely a ploy to save xAI.