# AI Infrastructure Economics and Market Discipline

**Podcast:** The AI Engineer Podcast
**Published:** 2026-06-05

## Transcript

Welcome to the podcast.
Kicking this off, the S&P 500 is refusing to change the rules for SpaceX to join the S&P 500.
This is also going to block OpenAI and Anthropic when they do their inevitable IPOs here coming up quickly from joining on a fast track.
We also have news that Google is going to be paying SpaceX $920 million a month to rent XAI compute capacity.
This is on top of what Anthropic is already doing.
Enterprises are blowing past their 2026 AI token budgets by 3x and a lot of them are trying to figure out how to put controls on this.
The CEO of Airbnb, Brian Chesky, is starting his own AI lab.
He was kind of was formerly advising Sam Altman and I think he's going to keep doing that but now he's going to be spinning up a lab which could be, you know, directly competing.
And speaking of competitors, about 90 investors currently hold stakes in both OpenAI and Anthropic.
I want to break down some of the numbers about why this is and why it is pretty uncommon for this type of thing to happen in Silicon Valley.
I wanted to mention, if you want to get all of these news stories in an email to your inbox every single day, go subscribe to the AI Chat newsletter.
It's on AIChatDaily.com, my own site.
You can click the subscribe tab in the top corner and you can subscribe to this newsletter.
It's free.
It goes out daily.
And I vet every...
story, what's going on.
I reorganize it and try to give you basically all the extra details that sometimes aren't covered in the podcast.
You get a deep dive or sometimes if you don't have time to listen to the podcast.
I'll leave a link in the description to AIChatDaily.com.
I'm blown away.
It has a 40% open rate.
People that listen to it love it.
We have listeners from all the top AI companies, so go check it out.
Link in the description.
Okay, let's talk about what's happening with the S&P 500.
They have refused to bend the rules for SpaceX, and of course, this is going to block opening Ionanthropic in the future too.
But basically, SpaceX went and they requested that they could skip kind of some of the standard rules for getting fast-tracked into the S&P 500, and the S&P 500 said no.
It's pretty obvious why SpaceX wanted it.
It would have basically given them 14 billion dollars in automatic passive fund buying just by getting included into this.
But the S&P actually has rules to get included.
One of them is a 12 month seasoning period.
So basically, once you IPO, it's got to wait 12 months before you can get added.
I guess this is just to make sure that companies don't just boom and bust and it doesn't have any sort of negative impact on people in the S&P 500.
There's also a 10% float requirement and there's a profitability screen that some of these huge IPOs, especially these AI models, don't really have right now, these AI companies.
So because of all of that, SpaceX didn't.
actually qualify.
And the idea here is that it's going to protect the $7.5 trillion fund, the S&P 500, right?
One of these mega funds from basically absorbing any sort of unprofitable or these companies that had these huge caps, but maybe they're not going to be successful long-term.
Now, is this some sort of indication that SpaceX isn't going to be successful long-term?
I don't think so, but I think they're just saying, look, we have these rules.
You got to follow them.
And while SpaceX is missing out on about $14 billion in automatic passive buying, OpenAI and Anthropic are both going to, you know, because they're not also going to get included, they're both going to miss out on $8 billion for OpenAI and about $4.6 billion for Anthropic, according to Bloomberg Intelligence, if they had kind of got in there early as well.
But I think because they're saying no to SpaceX, those other two are basically no's as well.
The NASDAQ 100 is going to let SpaceX in 15 trading days and FTSE.
Russell's Russell top 500 is going to let it in in five days, which is far faster than they normally do.
But these are in no way as big in size as the S&P 500.
So there's going to be smaller, but still it's going to get some passive inflows.
The reason why I think this matters is because S&P 500 right now, this is sort of an interesting move for index discipline, right?
They're saying, look, we're going to follow our rules over the capital flows.
They're missing out on a lot of money.
A lot of people.
may go and invest their money directly into SpaceX when they could have just bought the S&P 500 knowing that they would get a piece of SpaceX.
So now they have to go and take.
either their money out of the S&P 500 or put money that they might have put into the S&P 500 alternatively and put it directly into SpaceX to get exposure to this stock.
Specifically, as these companies are, you know, hitting these trillion dollar valuations, opening I Amthropic at SpaceX, all of those are not going to be in the S&P 500 for at least 12 months because of the rules.
And so if you want to get a piece of those, you've got to go direct.
So an interesting move by the S&P 500, but I do respect some of the discipline there.
Google is going to be paying SpaceX $920 million a month to rent XAI compute capacity.
This is to get access to 110,000 NVIDIA GPUs.
They're going to get this through June of 2029.
It's a 32-month deal.
And this is just being disclosed to SpaceX's prepping for their $1.75 trillion IPO, which is happening next week.
So we have this massive IPO coming up, and they've dropped this news on us.
Obviously, SpaceX is trying to increase their revenue as much as possible.
ahead of this IPO.
The deal is going to run starting in October this year.
So if they have a massive run, you know, they're training something huge with Grok right now, they could probably have it everything kind of done, buttoned up and moved on by the time this is happening.
A lot of people are saying, well, does this mean that, you know, SpaceX isn't getting the usage they hoped for Grok?
Either way, they're making a lot of money, so I think they're probably in a happy place.
But this is going to cover GPUs plus CPUs and memory components, and it's all going to be inside of SpaceX's data center, and this is going through 2026.
This is SpaceX's second huge compute deal.
Anthropic, like we know, they signed a $1.25 billion a month agreement last month for the entire Colossus 1 Memphis facility.
Google's Alphabet is facing $180 to $190 billion in 2026 CapEx.
They announced an $85 billion stock sale this week to help fund that.
I think they're showing that they're expecting even larger AI infrastructure costs in 2027.
So basically...
Everybody wants more compute and SpaceX had kind of an overbuild it would seem as they're kind of selling off to Anthropic and to Google.
And while you know, some people would have said like, hey, look, you guys are going to be in the hole for overbuilding.
Well, it turns out you can just turn around and sell it to all of your competitors who seem to have an endless demand for their AI.
This is particularly important when you look at the fact that SpaceX's AI unit is losing $2.5 billion a quarter and they have.
$2 billion in monthly contracted revenue locked in.
So I think they're going to try and make sure they can keep pushing that number up.
I would expect perhaps that we might even see more people signing deals to buy some of their compute.
All right.
Enterprises right now have blown past their 2026 AI token budgets.
They've done it by 3x.
A lot of them are scrambling for controls.
I've talked about Uber on a previous podcast.
Uber burned through their entire coding budget in four months for the whole year.
There's another company that is facing a 500.
a million dollar Claude bill after they failed to set usage limits for their developers.
I think this massive explosion right now is coming from a lot of these AI models that were released in November.
In November, it felt like Anthropic just had some really incredible models come out.
Personally, since then, I've been using, you know, I switched from OpenAI to Anthropic being my primary tool, and I built a ton of SaaS, a ton of apps, a ton of really cool projects.
It's a lot of fun, but you definitely can burn through an insane amount of usage, especially when Anthropic just released their work.
flows, which, you know, you can tell it like, hey, go do this really long, lengthy, complex task.
It can spin up dozens of agents below it and get them all running at the same time.
So you just burn through so many tokens.
Now, in my case, I'm using the CloudMax subscription, which is $200 a month.
But from and it's very heavily subsidized, I think, according to my calculations, but, you know, some people may disagree with me.
You can use.
somewhere between $5,000 and $11,000 a month in what would be Anthropics API costs if you absolutely max it out.
And so, yes, $200 a month, great deal.
But a lot of people don't have the max plans.
A lot of people are just grabbing directly from the API.
Their company is just letting them use something like GitHub that no longer has a subsidized plan.
You just have to buy the API tokens as they come.
And so in that situation, you can be spending an insane amount of money as you...
build things.
Now, the reason why this is important, Priceline's cursor contract renewal cost them four to five X more than they were expecting.
Their per developer token consumption rose 18.6 X in nine months across a bunch of different companies that they were tracking.
The top token users were two times more productive than light users, but they burned 10 times more tokens to achieve those gains.
According to Jellyfish.
research.
Now, I think, you know, some people would be like, well, you know, it's not even worth it because you might be two times more productive than light users, but you're, you know, using so many more tokens.
At the end of the day, the solution is getting the max plan, which is subsidized.
How long will that last is the real question.
And I don't think it can last forever because Anthropic can't subsidize this forever.
But in the meantime, if you are an organization, I 100% recommend not going through just, you know, directly using APIs.
You can cut.
You can cut costs dramatically.
The Linux Foundation launched the Tokenomics Foundation this week to standardize AI cost metrics with 180 vendors already participating.
So it seems like some other people are trying to tackle this problem and put a little transparency into this.
Airbnb's CEO, Brian Chesky.
He's going to be starting his own AI lab.
It's going to be focused on user interaction and design.
Everyone knows that Airbnb is famous for design.
Brian is famous for being kind of the head of design there.
He's obsessed with it.
It's something that I, you know, whenever I'm thinking about how to design the UI UX on an app on a tool, I always go and look at what Airbnb is doing and try to find some precedent there.
So very famous.
And this is what he's focusing on.
But he's actually going to keep his CEO role of Airbnb.
And he is going to be directly competing with open.
This is a company that he's advised for years.
He said he believes that the existing AI tools lack the rich interfaces needed for travel and e-commerce.
Which to be fair, I think that a lot of these AI tools are not great at design.
There's ways to get around it.
I think the best way that I've been using lately is just screenshotting a website that I like the design of, giving it to Claude and telling it to make it in that same style.
And it seems to do better.
But if you just ask it stock, like, hey, build a site, build a thing that does XYZ, the design is not going to be a beautiful site.
And so I think this is what Brian is seeing specifically for Airbnb.
He's not going to be running this new lab himself, though.
He's going to have another executive that's going to be the CEO.
and he is going to act as the founding chair.
It's basically kind of his hands-on founder mode approach at Airbnb.
Airbnb has rejected ChatGPT plugin partnerships because underlying tools aren't robust enough, according to Brian, but also this might set the stage for him to create a competing lab and use his own tool, kind of like how Elon Musk created XAI and plugged it straight into...
X and had Grok as kind of the big model there, you could probably make a model and get a lot of usage if Airbnb was using that directly.
Brad Adcock's Hark launched late last year, and I think it's doing something in making kind of a similar bet, focusing on the fact that AI models are not great at UI and UX.
About 90 investors hold stakes in both OpenAI and Anthropic.
Now, this is not something you see in Silicon Valley a lot.
People don't usually bet on two different competitors, but according to a new report, 42% of OpenAI backers also hold Anthropix stakes, which feels a lot like investors are trying to hedge their bets on this.
We have Sequoia Capital, Greylock Founders Fund, and Redpoint Ventures.
They all sit on both cap tables.
At least 13 of the 31 invested names in Anthropix latest raise already own OpenAI shares.
Each company has raised over $100 billion at, you know, basically a trillion dollar valuations ahead of their IPOs this year, close to a trillion dollars, you know, somewhere between 850, 950.
But they're raising so much money that I think a lot of the traditional conflict of interest type of governance concerns are basically gone.
These are massive raises.
Roughly 30% of the overlapping investors are hedge funds, private equity, or wealth managers.
These people are very used to spreading their bets across competitors.
Traditional venture firms, I think, are now starting to adopt that same strategy.
It feels like everybody wants a piece of every pie before the big IPO season is about to be honest.
Guys, thank you so much for tuning into the podcast.
If you haven't tried AI box out already, you can access 80 different AI models in one place for $8.99 a month.
This is my own startup.
You get access to all of the image, audio, text, and video models.
I would love for you to try it out and let me know what you think.
You can stop juggling tons of tabs with different AI models.
tons of subscriptions to different places.
You can get everything in one place and chat with all of them in the same chat thread.
It's super useful.
And the link is in the description at AI box.ai.
All right.
I'll catch you guys all in the next episode.