Invest Like the Best with Patrick O'Shaughnessy
00:02:22 · Patrick O'Shaughnessy, Ben Thompson
[00:02:22] S0: So, Ben, if you can believe it, how long it's been since we last did this, world was very different. No AI at the time. We talked about aggregation theory mostly. I thought a fun place to begin since the world has changed so much is to hear what you think it would mean for The US to win the AI race. [00:02:37] S2: I think it would be very problematic for The US to win. Let's say we take the most sort of fantastical scenario where if you control AI, you basically... Your military is better than anyone else. Somehow it fixes our manufacturing, all these things that I don't think AI is necessarily going to do because they sort of deal with the real world. But in this world, what is the game theory optimal response of China? To blow up TSMC, game theory can get very sort of convoluted and complex. To me, this one actually isn't that complicated. There's a fundamental disconnect that I have with a lot of the rhetoric coming out of Silicon Valley, coming out, I think, one of the labs in particular, where if we get to a place where we have a meaningful superiority in terms of a military national security perspective, I think that's very dangerous for the world. [00:03:28] S0: But in that state, how much does it extend beyond TSMC being blown up? Because in that state, I would assume we figured out how to build fabs here in The US, you know, to some degree and are less reliant on that one choke point.
00:03:40 · Ben Thompson, Patrick O'Shaughnessy
[00:03:40] S2: I think there's a little bit of magical thinking, which I just invoked in terms of manufacturing and whether it be fabs, whether that be actuators, all these precursors. I think the degree to which we are dependent on China is underappreciated and is not something that is going to be fixed outside of a conflict. Just because fixing so many of these things is going to be dramatically dumb. If your competitor is sourcing from China and you're gonna start sourcing or getting things from The US, you're gonna be at such a disadvantage relatively speaking that you're just not gonna do it. So you do it when you have literally no choice. And that works for very big headline items. Like you can browbeat Apple to move some of their iPhone manufacturing to India, for example. But even that is a good example because Apple is not truly moving out of China. They're diversifying to an extent, but it would just cost so much. It's like paying an insurance policy that if you don't have to pay it and it's astronomically expensive, you're just not going to pay it. It's one of those sort of hypotheses that I just have a hard time even grokking because the only world I see where we truly pull out and have no dependency on China, such that if they want to blow up Taiwan, who cares? There's no impact on us. It seems pretty fantastical to me. And I think there's a bit of facing reality in this regard that is not present in [00:05:12] S0: these conversations. Put yourself in their shoes. Like what do you think the motivations are? Everyone
00:05:12 · Patrick O'Shaughnessy, Ben Thompson
[00:05:12] S0: these conversations. Put yourself in their shoes. Like what do you think the motivations are? Everyone [00:05:17] S2: can use a good bogeyman. I think from the AI trade perspective, nothing works better than we have to beat China. And I do think we need to beat China. We need to be competitive. I despair at the extent to which over the last few years in particular, so many of our responses, particularly from a political perspective, has been to like try to be like China. I think we should be going the other direction. More openness, more innovation, less top down control, less restrictions on speech and things along those lines. America succeeds by being on the leading edge and by leading into that. [00:05:54] S0: You said probably The US being purely dominant in AI is not the right end state for the world. What is your ideal equilibrium for how this goes worldwide?
00:06:03 · Ben Thompson
[00:06:03] S2: There's a bit where AI right now is kinda like the Taiwan situation. In that the current status quo actually doesn't seem so bad. The question is how sustainable is it? But maybe it's sustainable for longer than we think. The way I think about it right now is I think OpenAI and Anthropic are clearly on the frontier. Who knows what's happening with Google and then Grok and Meta are chasing them. Meanwhile, the Chinese are very capable, very smart, and also definitely distilling these models to sort of stay about six to nine months behind. And it feels like a pretty good equilibrium that I think is generally favorable to The US. Now the question is how long could it stay this way? And there's a lots of questions out there. Like, can the Chinese actually pull ahead? I'm still a little skeptical for various reasons, whether it from chips. Getting to the waiting edge, I think that last six to nine months is very difficult. It's gonna be instructive how Meta and Grok do in terms of actually catching up, especially as we get to the world of AI improving itself, using AI to make the AI better, which I think is definitely a real thing. I think you see a real acceleration from both OpenAI and Anthropic recently, which was sort of theorized. It seems to be coming true. And to the extent that's true, can you actually catch up? And I think the other question about this, by the way, is to what extent does that apply to cost to serve, to marginal costs? If you can apply AI to optimizing your stack, to figuring things out, to analyzing all the data, is your cost to serve structurally lower than anyone else? This is the thing about the open source models. The talk about them being free is bizarre to me because it's marginal costs. You still have to run inference like GLM or Kimi. Kimi is very expensive to serve. The cost per answer is significantly higher. Everyone referring to these as free. It feels like in the narrative, it's in people's head that free is free. Now I can use AI for free. No, you can't use AI for free. You're not paying necessarily the R and D to create the AI, but you're definitely paying the inference to sort of run it. So right now, I kind of like where we are and the pushback would be that's right now. It's not gonna stay that way, which I think is fair pushback, but I don't know. This way longer than we think.
00:08:16 · Patrick O'Shaughnessy, Ben Thompson
[00:08:16] S0: If you can know anything about the future of how this will go to be more confident in, like, where the equilibrium will end up, what is it? Is it like the length of the s curve? Like, how far up the s curve we are? At some point, these things presumably will level out, maybe not. What would be the thing you'd wanna know that would give you a better sense of what the future might look like? [00:08:34] S2: I am concerned that with the scare around people freaking out about mythos and this hugging face incident, that the actual implication of that is not that we reduce these dangers, but we just stop releasing stuff. We on the outside start to lose any sense of like where exactly What's [00:08:58] S0: actually the frontier, yeah.
00:08:58 · Patrick O'Shaughnessy, Ben Thompson
[00:08:58] S0: actually the frontier, yeah. [00:08:59] S2: What is actually the frontier and where it is. There becomes sort of a false sense of security because right now everyone's basing their understanding of mythos on fable. But how good is fable actually relative to mythos? That sort of gap is only going to, I think, increase over time. So I think that's a real question that I'm not sure about. This question of the recursiveness and AI sort of making itself better, does that lead to some sort of takeoff? And at the end of the day, there's timing questions in lots of different ways. I'm worried about the timing mismatch in terms of the actual return on investment producing enough revenue to fuel investment. We're working our way down the capital curve. We started with free cash flow. The speed with which the tech companies blew through the debt markets is kind of incredible. It took like a year. And now Google's issuing equity. Nvidia's putting together these [00:09:51] S0: This $500,000,000,000 thing. This $500,000,000,000