Invest Like the Best with Patrick O'Shaughnessy
00:30:14 · Patrick O'Shaughnessy, Ben Thompson
[00:30:14] S0: I know you've been spending time with some of the big money firms and sources of capital. What is your sense of their appetite right now and how they're thinking about the future? Because I think this year, it's gonna be 800,000,000,000 or something that we're gonna spend in CapEx. Next year is supposed to be 1,300,000,000,000, I think, the current estimate. It's gonna keep going up from there. We're burning through all the compute that gets installed basically immediately. It's it's such a strange circumstance that we can use the capacity right away as soon as it's online. [00:30:41] S2: Well, that's the thing, though. So there's a few timing mismatches that are happening right now. We can't use it right away. All the bulls on Twitter is always like, we don't have enough compute. We don't have compute. Well, we don't have enough compute because there was insufficient investment made in 2023 and 2024, which yes, absolutely. And by the way, if you think there's not enough compute, TSMC decreased their rate of growth in 2023 and 2024 and 2025. Our shortage of compute is going to get worse in the next few years because a fab, the lead time is even greater than a data center. Today, when we say there's not enough compute, it's not like all the money that the companies are putting [00:31:19] S0: in
00:31:19 · Ben Thompson
[00:31:19] S2: today manifests Yeah. In No. It all manifests in compute in 2028 and 2029. On the calls, have both Andy Jassy and Syna Della are out there saying, look, we're just building data centers like these are the shells. We might not use them now. Maybe we'll use them in the future. And we only buy GPUs when we know there's demand for them. That is a great story to tell. I'm not sure that I think is a lot of BS because the reality is if you've built the shell, that money is sitting there. You're not gonna let it just sit there. If you invested a fixed cost, and this is the whole logic of commodity markets. I think tech in general doesn't understand commodity markets. Tech is by and large focused on if I produce a highly differentiated product and that differentiation could be like software, it could be a network in terms of developers, It could be a social network sort of thing where peer to peer, where I'm highly differentiated than my ability to charge higher prices provides sort of my profit margin. So the classic example is like Apple. They have their ecosystem and they have their software and they have third party and all those sorts of things. And so they can charge 50% margins on their iPhone. Everyone looks at Apple as like the ideal business model. That's how you run a business. But in a commodity market, the price is set by the marginal supplier. Cost to serve is all that matter. That's right. I had a good friend in Taiwan who is in shipping. Fascinating industry. It's kind of like the airlines too. Another industry that I love to look at. You buy a ship and the cost of that ship is depreciation. Your marginal cost is actually quite low. It's the fuel to run the ship and the cost of the crew and like your port fees. Not that much. What that means is you are going to run that ship- As full
00:33:01 · Patrick O'Shaughnessy
[00:33:01] S0: as he
00:33:01 · Ben Thompson
[00:33:01] S2: really No, you're run no matter what. And you're gonna bring down the price of a container as low as it needs to be to cover your marginal costs. Now your paper losses in this situation might be very large because your accounting loss includes depreciation. But the depreciation is an accounting segment. You already paid the money. You're going to run that ship at whatever the market will bear in the container. The beauty of the container, it is a pure commodity. The cost of the market is gonna be the marginal cost. Now, if it gets low enough at some point, people will exit because their marginal costs, they're actually losing money on a shipment, not just paper money, but like actual real money. They will exit, but then the supply is diminished. So then the price will go back up and you get this interplay of sort of coming in and off. But then let's say the market's very high like it was during COVID. It's like, wow. We're making so much money right now because there's not enough supply. There wasn't enough supply of ships. So containers went from usually being like 3,004 thousand dollars to 17,018 thousand dollars. The amount of money that these shipping companies made in a very short amount of time was insane. What happens though? Well The more ships. Imagine if we had more ships. Right? The problem is it takes two years to build a ship. If everyone makes this decision simultaneously, you suddenly have a lot of ships, price plummets, etcetera. Where we see this is in components, in memory in particular. Memory, very famous for boom and bust cycles, people entering the market late. But to what extent are data centers going to be memory makers? Where right now everyone can see we don't have enough compute. So everyone's like, we absolutely have to be investing because there's so much money to make. And look at our payback period. The problem is you're measuring your payback period in a time of scarcity. Is that payback period gonna hold in a time of abundance? And the sort of the bulls say, there's never gonna be a time of abundance. AI We're gonna be short forever. Scaling. We're gonna be short forever, which maybe we will be.
00:34:57 · Ben Thompson, Patrick O'Shaughnessy
My concern is even if that's right, we could still have an air gap in that there's so much money going into it right now and not enough has come online to actually make sufficient revenues to handle the situation where we run out of capital. I believe in AI. I think it's a real thing. I think the economic impact is gonna be astronomical. I think all the concerns about societal impact are very real and are going to come to bear in a major way. You can believe all that and still be worried about, are we gonna make the bridge to this actually generating the level of returns necessary to continue to fuel this sort of going forward? [00:35:31] S0: Can you zoom in on TSMC and the component makers where fabs are involved? And so far, at least my understanding is that they've been quite conservative in their willingness to expand capacity, build new fabs, meet the market's demand with similar growth, which they have not done. If that just rate limits this whole thing and prevents us from getting one of these giant overbuilds?
00:35:52 · Ben Thompson
[00:35:52] S2: We can talk about a few different ones. Like, we'll start with memory. Memory used to have tons and tons of memory makers. Every time there'd be a boom, memory makers would sort of reenter the market. New countries would come in like Taiwan used to have like a memory market, but you would get these exact dynamics. If there's a shortage of memory, there's so much money to be made. You can't bring capacity on immediately. It's the same as shipping, the same as what we're seeing right now. That would spur people to come in the market. You'd get too much capacity, prices would plunge and people would just get blown out. Because the issue is the upfront cost for these is so large, just like buying a ship, like building a fab is even more so. And memory now, like the leading edges of memory are using things like EUV machines. So the costs are getting into the billions of dollars for these lines. What happens is every time with these boom and bust cycles, some people would enter, more people get washed out. You go through these famous historical moments for these memory cycles. Companies just get blown out. One of the most interesting actually memory stories is how Samsung sort of took over memory was they saw it as an opportunity and they had studied history and they realized that actually the way to take over the market is to invest into downturns so that you're ready when the next cycle comes around, which requires a ton of guts and a ton of discipline and a ton of money. But they did that and basically wiped out the Japanese. That's when the South Koreans generally took over the market in a major way. But it got down to three. And the problem is three, it's not a monopoly, but it's kind of an oligopoly. And they all got a lot more discipline about let's not make the mistakes of the past. And we're not colluding, but we all are on the same page about let's not do that. And I think that dynamic sort of ran head on to the current moment where it just took a while for them to realize, no, there is a secular shift in memory demand that didn't exist for a very long time. I think the memory solution will be solved eventually. The other risk they run is Apple's lobbying to get Chinese memory. What is the number one focus of like arrhythmic changes? How can we use less memory?