Carlos E. Perez says the deal mimics AWS.
Positive users see xAI selling compute to Google and Anthropic as a shrewd flex and rational infrastructure play, while negative users call it proof xAI has lost and is retreating from the frontier race.
okay @kateconger with the important context, google (prob to calm its shareholders) says this is *short term*
so p hard to determine whether this will even be real ARR or just a nice bonus booking to see before the IPO (but doesnt start till october)
https://www.nytimes.com/2026/06/05/technology/spacex-google-deal.html?unlocked_article_code=1.oFA.pwqq.wQrMgdMrNXmI&smid=url-share
@MikeIsaac SpaceX is a bet that scaling laws are very long and there are reasons why one would want to build the Dyson Sphere.
If they manage to say reasonably behind; they nonetheless win by default, as they'll be the only ones building the Dyson Sphere.
I am skeptical
@MikeIsaac elon companies don’t give up. the market is still young
@TheMattBass @MikeIsaac SpaceX is leasing their old data center cluster Colossus1 to Anthropic. Xai is already being trained with the Colossus2 cluster.
@MikeIsaac They probably are waiting for Cursor deal to close to revamp/re-ramp new AI model team, so in the meantime rent out unused capacity after nearly all the original xAI team left (notice the easy opt out terms if SpaceX gets internal AI traction)
@MikeIsaac It’s a sign for sure that they don’t need the capacity
@M_Cottone not going to touch all the data centers in space stuff BUT i do think it is smart to rent unused capacity and drop it asap before listing
how long it stays on the books is another q. but it's not a bad move for today at least
IMO this is a very smart decision by SpaceX with the IPO coming up. Like it’s crazy, don’t get me wrong, and Elon’s ego is still going to drive him probably to develop/claim Grok as a real competitor, but if his long term play is data centers in space/total world domination, seems like a good way to do it would let the other model companies duke it out and pay you while they do it.
That is unless a singleton ASI model takes over, which is probably the only super downside of this from SpaceX standpoint.
But that’s an outlandish type scenario not really worth considering if you’re thinking about mortal things like ARR.
And/or if the “bubble” pops obviously.
Yeah I mean the data centers space is yadda-yadda-yaddaing a lot of materials science, rocketry, AI engineering stuff. But when analysts are telling you, the richest man in the world, that your company’s revenue is going to be the size of California’s entire economy in 15 years, you probably think anything is possible
Doubt it, these contracts are short term / have an out. What it means is he is building up a massive compute capability where he will be able to scale it enormously whenever needed, and he can have it generate profits in the meantime. He will make more when they are running his models though, so the incentive is always there to keep trying.
If you believe in AI, owning the compute layer is really smart. Owning the power/real estate layer is even smarter, hence the space stuff.
Ultimately, power and mass are the spice.
If you remember Dario’s interview with Dwarkesh, he’d very clearly said that there is a split between inference capacity (for revenue) and capacity that’s used for R&D. I think XAI inference part of their capacity is basically what is being given out to Anthropic and Google - let’s say that’s 70% or whatever and the remaining 30% is what’s needed for them to do their R&D. so I think that could still be the case with XAI.
@GaryMarcus spacex + google collab amazing fr
@GaryMarcus not wrong tbh. lending your edge out usually means u dont trust ur own speed anymore
@MikeIsaac my read: the Grok team got torn to studs and is being retooled under the stewardship of Cursor and that eventual acquisition. It's a short term lull but not a cessation of model development. Might as well sell some compute in the interim
@MikeIsaac Didn’t xai also just sign a deal with anthropic as a sign they’re giving up on their crap modeling being so far behind the others? Hence his rolling all his failures up in one ipo?
@MikeIsaac half empty view is that Google owns $100b of the inflated equity and the deal is cancellable on 90 days' notice. It is like dark fiber with an equity kicker.
@GaryMarcus Yes, and? You’re mad because they’re pivoting to something that makes more sense?
@firstadopter @MikeIsaac Cursor’s Conposer model has proven to be a very good coding model. XAI will use that and make it better with Collosus capacity. I suspect they don’t need high utilization (yet), so they are renting out the excess capacity.
@MikeIsaac Nah, the've got the option on cursor and will use them to develop a frontier coding model. Cursor's issue is capital and spacex has it. they'll catch up quick
@GaryMarcus So what (if even true)
SpaceX is about so much more than this drop of water in the ocean
