alt Hacker NewsGoogle to pay SpaceX $920M a month for compute capacity at xAI data centers
Comments
This is a masterful piece of financial engineering by Google and SpaceX.
Google purchased 10% of SpaceX over a decade ago. After dilution they probably own around 5%.
SpaceX is valued at a whopping 94x revenue. This deal increases SpaceX's revenue by $11 billion per year. If SpaceX maintains this revenue multiplier, then this single deal boosts SpaceX's valuation by 94 x 11 billion = $1 trillion dollars. Google owns 5% of SpaceX, so they make 50 billion dollars. Google spends 10 billion and makes 50 billion, $40 billion profit.
The even better part is that because of this deal, SpaceX is now profitable. The S&P requires companies to demonstrate 12 months of profits before they can enter the S&P 500 index. SpaceX lobbied to have this profitability requirement removed, but S&P said no and refused to rewrite the rules.
Now with this incredible deal, SpaceX is now GAAP profitable under the existing rules, and they get to join the index next year without a rule change.
Truly a brilliant deal for everyone involved.
And SpaceX will spend $800M per month on Nvidia hardware purchase contacts, and Nvidia will spend $700M per month on Google services.
I'm picturing a teenager blowing a bubble gum bubble bigger and bigger. I assume it can go on forever!
Google renting infra from xAI, I did not see that coming. My understanding of what computers are doing, what companies are doing and what governments are doing seems to be getting worse day by day.
A huge chunk of SoaceX value in their filing is attributed to their AI technology (aka Grok). I believe it’s 90% or more… Now, it seems they’re leasing the infrastructure required for Grok to scale to Anthropic and Google. I wonder how that math works…
google: we are commited to carbon free data centers by 2030 (https://sustainability.google/reports/247-carbon-free-energy...)
also google: renting capacity from a data center powered by 27 methane gas turbines on trailers
https://www.epw.senate.gov/public/index.cfm/2026/4/whitehous...
> $920 million per month from October 2026 through June 2029 for access to “approximately 110,000 NVIDIA GPUs, CPUs, memory, and other related components.
That's about $8,400/month per "component" is that in the ballpark at all with what a month of dedicated/exclusive access to an NVIDIA GPU would go for?
> compute from its Colossus 1 data center near Memphis, Tennessee, that xAI — now part of SpaceX — originally built for its own artificial intelligence efforts.
Is this the same Memphis data center notorious for burning jet fuel nonstop for power?
You guys don't understand. Banks like JPMC will make billions on this IPO. Doing everything to prop it
Is there any data on whether Google, Amazon, Microsoft, Anthropic, OpenAI etc are most cost efficient in getting datacenter compute online and operating it?
I'd be interested in how large the range is here across company and region and specific data center and how it relates to companies like Hetzner if at all.
These deals are part of how the AI economy operates. Amodei has explained this in his recent Patel podcast.
1. Building datacenters takes time. Months, if not years. They take billions of investment.
2. AI revenue is highly unpredictable. Sure, you can make predictions, but maybe your competitor releases a better model 2 weeks after your release, maybe the new model you built isn't as much better, maybe the chinese models steal your show, etc.
3. AI revenue grows a lot. Anthropic's case is 10x per year.
4. So if you are off by just a year in terms of how much GPU you actually need, then that means a 90% of your compute capacity is wasted, and you go bankrupt.
As a solution, companies buy compute from each other! If one company's model did well, they can buy compute from the company whose model didn't do well (like in the case of grok). It's beneficial for both sides, so positive sum game. So deals like this aren't something bad in itself.
It's nothing new either. In SAAS deals, you often commit to a certain revenue and then pay extra if your revenue exceeds that amount. And power market is cut in two as well: longer term deals plus spot markets. Spot prices are way higher than the longer term deal prices.
Given it's SpaceX of course there is financial engineering involved: the GPUs aren't actually owned by SpaceX but a daughter company, and it's been financed via loans that are backed by pension funds. So it's already the case that pension funds back bear the risks associated with SpaceX's operations.
Right now, the bulk of the AI bubble sits in such debt statements and not in public markets.
380 dollars per second... Good to know I could afford my own data center for an appreciable fraction of a minute.
Google has code written for their Tensor processors (TPU). Will it run on the NVidia GPUs that xAI has? Because I'm thinking they're "not part of their core architecture" and it will thus be money wasted.
(I thought for sure the title was backwards - it's a strange world)
> If SpaceX maintains this revenue multiplier
Yeah, if a ridiculous premise is given you'll reach a ridiculous result.
All important technologic revolution can turn into an infrastructure revolution => the railroads, electricity etc … so the technology can be a transformation while the capital cycle becomes in excess
> SpaceX said in the filing that if it fails to “deliver access to the committed amount of GPUs by September 30, 2026,” Google can immediately end the agreement, or accept the number of GPUs provided at a reduced fee after a one-month grace period.
> After this year, the agreement can be terminated by either party provided they give 90 days’ notice.
Circular financing at its peak for the IPO. There has to be some regulatory body to not allow such shady things
“If we fail to deliver access to the committed amount of GPUs by September 30, 2026, then following a one-month grace period, Google may immediately terminate the agreement or accept the number of GPUs provided, with a corresponding pro rata reduction in the monthly fees. After December 31, 2026, the agreement may be terminated by either party upon 90 days' notice.”
https://www.sec.gov/Archives/edgar/data/1181412/000162828026...
It’s only to boost the IPO price. The agreement will last only a few months on paper. I doubt it is a real transaction.
I am really wondering if Google is just subleasing Anthropic capacity. The terms are suspiciously the same as Anthropic's and Anthropic was supposed to have leased out all of Colossus 1.
Maybe with the corp token spending limits and the rise of codex, Anthropic saw steep deceleration in usuage?
Is this admission that google’s proprietary chips etc. are not cutting it? Why would you need a bunch of nvidia GPUs if you have your own silicon? (AFAIK they have their own for both inference and training do they not?)
Do people think those numbers are correct? 920 million a month for 110,000 GPUs is $11.61 per GPU hour. That seems very high to me.
With ~$2.1B/month of GPU rental revenue, is xAI now profitable? Are all divisions of SpaceX now profitable?
Which means more Grok degradation, more severe throttling, etc.
I can't understand why xAI charges 50% more per month for Grok over competitors when it doesn't even gracefully downgrade to a cheaper model when paid subscribers hit the limit.
That’s a lot of moolah.
Makes a lot of sense that Musk should do the parts of the AI stack that look more like manufacturing/regulatory bottlenecks, and rent out the compute to research-focused AI labs. Does anyone know the full accounting of how much it cost to build Colossus (plus ongoing opex) vs. the revenue it's generating now?
I guess their training runs aren't going great if they're dumping all the compute they can.
Or I guess juicing the numbers for IPO
Tangent alert: a couple of questions for folks who know far more than I do about compute capacity and Google these days...
Lately, like the past few months, I've noticed Google services (search, gmail, drive, maps) running very slowly to the point where, at the moment it happens, I always think it has to be my connection and not Google, but sure enough every time I check a couple of speed tests and they're... fine. And then I don't seem to be having the same latency from other sites/apps. Is there any chance that the commingling of the AI snippet and then directing users into the AI funnel through the text box is actually causing material performance impacts in other Google properties? Probably a dumb question because I can't imagine they would allow performance for broader properties to suffer for AI prompts/chats, but then again all this talk of compute starts making me think otherwise, like the prolific amount of prompting and chatting is causing massive across-the-board performance issues.
Somewhat related, but does anyone use Gemini and end up with the experience where you have a chat and it's obvious, to yourself and to Gemini, that you're trying to find a product to purchase, but Gemini doesn't even link you to what you would think would be the obvious place to purchase the product? This happens daily where I interact with it, it suggests some products, but won't even provide a link to that product or, if it does provide a link, it's to some no name site that wouldn't come up as a highly-ranked paid or organic result through regular Google search. Keeps making me think this is a Google performance problem where they have not figured out how to take the entire AI chat and engineer it back into a simple short keyword phrase to get an acceptable search result.
Btw, if anyone's thinking "why are you using Gemini because it's the worst?" I think that's fair and right. I have... reasons, but they're not super sensible ones.
I serious doubt Google is doing this for the spare datacenter capacity.
This is a ridiculous amount of money.
Have to believe a non-tech company could hire an entire team/company to build datacenters for this kind of money.
Make no mistake - this has to be “do evil” territory.
So Google AI will now be running partly on xAI data centers which run primarily on natural gas burned on site next to poor neighborhoods in Tennessee and Mississippi causing massive air pollution to these families and children. Is anyone else disgusted by this? I’m imagining all the people there developing lung and other issues because of this. Greed and power on full display over doing the right thing.
I’ll be switching off the Gemini model at work (Composer’s been off since their xAI deal). This is the final straw for me to move completely off Google services.
The Ponzi scheme that Elon has set up is one for the books. Fail forward. Unbelievable skill.
"It will have to be paid for," they said. "It isn't natural, and trouble will come of it!"
Fellowship of the Ring.
the GPU builds are very high stakes games of depreciation: if the mission life is e.g. 4 years you win, if a disrupting ASIC for the transformer comes in you lose.
As of today the gamblers seem to win, demand even for A100s, H100s is high prices are even rising.
When as appears inevitable Google decides to stop using this capability what will it do to the SpaceX stock value?
I'm curious if this offer lasts until after the IPO.
How can SpaceX have so much GPU spare capacity? It doesn't make any sense.
Did Musk blindly order humongous amounts of GPUs years ago before any of us had any sense of the scale this was going to reach?
Google owns 5% of SpaceX fyi.
If you can't buy DRAM, you gotta rent your compute infrastructure.
Sorry, what?
Does this mean that SpaceX are the only company that really did build some datacenters to put all the million of GPU/TPU/whatever they all talk about everyday?
I mean, Google, Amazon, Meta and Microsoft told investors they spent more than $1B per day last year in CapEx... why on Earth do they (well, Google and Anthropic at least) need to rent compute to SpaceX, of all companies?
SpaceX valuation and ultimate success depends on two things:
1. AI demand continues to grow. 2. SpaceX's orbital data centers are profitable.
If both of those are true, then their current valuation is absolutely justified. I'm confident #1 will happen.
#2 is the big bet, and IMHO this is just an engineering/execution problem. All they need is (a) Starship to work reliably, and (b) a manufacturing line that can build a data center satellite at low cost.[1]
(a) is the harder of the two, IMHO, but they are well on their way. Once they successfully recover and refly a Starship upper-stage, they will iterate step-by-step until launch costs drop to the level they need.
Now assume that SpaceX succeeds. What if AI demand continues to grow and SpaceX orbital data centers are profitable? Think of their moat: they spent 10 years and billions of dollars developing a fully reusable rocket that happens to also be the largest rocket in the world, and that costs 1/10th of what other rockets cost (per kilo to orbit). Plus, they have an assembly line that can build data center satellites cheaply, and they start fabbing their own AI chips.
How is anyone going to compete with that? There are a bunch of data-center-in-space startups, but none have their own rocket--they're going to have to pay SpaceX to launch them. Blue Origin is developing a rocket as large as Starship, but it's not fully reusable--they will never get the cost down to Starship levels.
What's interesting is that all the AI companies, OpenAI, Anthropic, and even Microsoft and Google, are mostly leasing their data centers from someone else. They think compute is a commodity and the value is the trained model. But if SpaceX has the cheapest data center with the most capacity, they will be able to extract profits from the AI companies or (why not) compete against them with their own model (Grok).
In 10 years we'll see whether SpaceX succeeds or fails. If they fail at this, they will retrench back to a launch company (assuming they are still in business). But if they succeed, they will be a massive company, and the synergy between their businesses will be so obvious that everyone will say, "of course they succeeded!"
----------------
[1] Don't be distracted by claims that "cooling in space is hard" or "radiation is a deal-breaker". Neither of those are insurmountable problems--they are just engineering problems. Crucially, they are problems that are easily solved by getting mass to space. If you can get mass to space cheap enough, those two problems are trivial to solve.
I knew GCP was third banana but what is even happening?
Space-X is an AI/datacenter company that also makes rockets
looks like elon web services (EWS) is the master plan all along :D
Even Google does not use GCP…
Wait, are these the same GPUs that were diverted from Tesla into xAI a couple of years ago?
This feels actually like a pretty safe bet for Google, they secure the compute in case it works (I doubt that the described volume will be available in the near future), while if SpaceX doesn’t manage to provide there is not much loss. I see it more as another way of blowing up SpaceX valuation on paper…
Because SpaceX has excess capacity.
We're all going to end up as Elon's serfs, aren't we?
Cloud companies were made to sell others compute. Now, one is buying billions of compute from SpaceX, a rocket company. That sounds so backwards lol.
Great work by Musk and his companies to be in a position to sell billions to cloud vendors. I'd have probably missed that opportunity while trying to build great rockets or AI models.
So space data centers are absolutely possible then. I heard a lot of skepticism about the feasibility but it looks like Google and Anthropic looked at SpaceX and trusted them to deliver on the promise and even signed deals worth billions.
Since the S-1 filing, xAI has taken over and is likely the largest share of revenue. I would estimate that ~95%+ of xAI revenue, and 100% of its profit, is from renting their datacenters.
This is a datacenter REIT bolted onto a social media company bolted onto launch business bolted onto a niche ISP. The expected price to sales is ~100x. The best datacenter REITs trade at ~10x and pay a dividend, which SpaceX does not. Meta trades at ~7x sales. Comcast is one of the best-run ISPs, and it pays a 5.5% dividend on a stock trading at < 1x sales.
To say SpaceX is overvalued is to even beginning to convey the magnitude of the situation. It's going to be very painful when the valuation normalizes.