Elon Musk's $1 Trillion Chip Factory | The Brainstorm 147

**Brett** (0:00)
I think that the great IRR in AI is going to dry up financing for these white elephants. As in, who is going to give the state of California or even the federal government money to waste on something when you have such better return on capital projects that are just sitting out there, begging to be funded?

**Sam** (0:17)
$100 billion into Louisiana for thousands and thousands of Starship launches annually, and Figure Releasing Index, the largest database of robot training data. Brett, let's start. You put out a great piece and tweets on how monumental this investment into Louisiana is.
What's the right way to think about it?

**Brett** (0:46)
Yeah. Relative to history, this is one of the largest infrastructure projects of all time, so you really have to go back to the US highway system. The US highway system was like $700 billion plus in 2026 dollars. This ranks ahead of the giant hydropower dam projects in Brazil and China in terms of scale of its ambitions. The International Space Station, adjusted for real dollars in a sectorally comparable scenario is $200 billion, so not quite that big. But the real question is, it's like, why are they building all of this stuff? This is like a huge slug of money to go into Louisiana. And one, there's a useful reason to be there geographically. To launch for AI satellites, you basically need to go due south. And so you need a launch location that gives you a stretch of water over which to launch due south. In this case, the Gulf of America, formerly known as Gulf of Mexico.
And Louisiana has a lot of natural gas and you need the natural gas to convert into methane to fuel the rocket. So actually the scale of the launch they're trying to do requires a lot of raw materials.
And three, you know, SpaceX has such, as we model it, extraordinarily high return on capital potential for its stacked businesses. Initially for Starlink, where we think return on capital for like the 10th Starlink rocket launch will be 100% annualized, and the return on capital for the 100th AI satellite will not be as good as that. It will be like high 20s, but that's because we think they're going to use most of that capability for internal R&D to get themselves caught up to the frontier. But by the 1000th AI satellite launch, we think returns will exceed those of Starlink. Then they're no longer a capital constraint because capital will flow into the business to try to fund these opportunities. They're launch capacity constrained, or they're basically launch and manufacture and all of the things that would get in the way from them continuing to grow constrained. This is one of many very, very large chips they're putting on the map of the US to try to address and dissolve those constraints to grow.

**Sam** (3:18)
I did like in the chart, the California rail, the quarter mile of track leading to nowhere is an estimated $125 billion project. So ROIC, extremely important when putting hundreds of billions of dollars on the table.

**Brett** (3:35)
You know, and it's actually, I think that the great IRR in AI is going to dry up financing for these white elephants. As in, you know, who is going to give the state of California or even the federal government money to waste on something when you have such better return on capital projects that are just sitting out there begging to be funded? And so one of the things I was posting about over the weekend is the idea that the AI and the growth in AI, the amazing economics of AI are going to pull a lot of businesses don't seem to be contra AI exposed off sides because their ability to finance is going to dry up. As in the data centers, Tesla's IRR on its terrestrial, sorry, SpaceX's IRR on its terrestrial data centers looks like it's on the order of 75%, meaning they can invest $30 billion in building a data center and net of all of the costs. The return on that 30 is like 1.75 times 30 going out compounded.
It's an amazing business. Clearly, you want to have capital attached to that kind of investment.
If you are over there getting a puny 10% IRR on another project, you'd rather have it there. Then to finance it, SpaceX might pay a higher rate, and that's how they'll suck the money out of those projects, including maybe the government borrowing rate has to go up, accommodating the fact that there's such obviously much better returns in a sector that can soak up huge hundreds of billions leading into trillions of dollars in capital.