A short conversational clip (Jason M. Lemkin with a second speaker) debating whether AI investment will exhaust available capital. Lemkin raises the idea of finite capital and ethical trade-offs in fundraising; the other speaker situates AI capex as ~1.2% of GDP—slightly above 1999–2000 bandwidth capex at 1.1%—and far below the 1840s railroad peak (~6% of GDP), noting differences in asset longevity.
The bigger issue is when will all the capital on Earth and Mars be exhausted? And I actually think the meta issue, you know, the CEO of Anthropic had this thing this week where he was explaining because they're supposed to be the good guys, right? This, we left OpenAI to build something safer. And he's explaining why we have to take money from the Middle East. Now, I don't know if you saw this, right? He's like, we have no choice. His point, I guess the flip side is he didn't want to, that this was lowering their, their, their pristine ethical values to take money from the Middle East. But his point is we have no choice, right? I think the bigger issue isn't 800 billion because people are invested. The bigger issue is when will all the capital on Earth and Mars be exhausted? And there is. That is a finite number, right? It is a finite number. You know, you know, Trump has already been lassoed into Stargate with David Sack. I mean, Rory can help me figure this out. There is at some point OpenAI could consume all available capital available to it in the Western world or in the Eastern world.
Let's talk about a more nuanced version there, because I do. It's actually an interesting question and I think Paul Kadrovsky did a piece, right, which is just looking at the capex investment in AI as a percentage of GDP and comparing it to two other booms, the dot com boom and the railway boom in the 1840s of that. Because again, I was trying to think through what are the comparables and those are. And the railway track analogy is a good one. So we're running now at about 1.2% of GDP, which is basically 1.2% of GDP is going into making CapEx data centers, which is a little above the 1.1% of GDP that was going into bandwidth at the top of the 99, 2000. So we were already slightly bigger than the bandwidth boom. And we know where that ended. Interestingly, and I did not know this, it was 6% of GDP in the railroad era a year. So at peak railroad mania, it went to 6%. Now, as they pointed out, in that case you were building an asset that lasted 150 years and in fact just today did a huge acquisition. Whereas in this case you're building an asset that appreciates over three years. So it's not like with like. But just to kind of give a sense of the outer bounds to your point, Jason, of how crazy could this get, right? We're already a bigger boom in terms of capex than 99 2000, but we're nowhere near as big as boom as the wheel went boom. The pleasing thing about that answer is it allows both parties to continue to feel great. If you're a nervous guy, like I sometimes I'd be like, shit, this is crazier than 99, 2000. Makes me nervous. The Internet was big. Dunno. If you're an optimist, you're like, hey, we're only spending one quarter of what we spend to build the railroads, and intelligence is better than railroads, surely, so we're fine.
