Invest Like the Best with Patrick O'Shaughnessy
00:40:11 · Ben Thompson
The entire fabulous model is so critical to what tech is and what it does. And also just the guts to do that at that time, particularly in someone who lived there, a culture that doesn't necessarily tend to make those sorts of bets. TSMC, they were pretty conservative, to be totally honest. So what happens though? Where'd the risk go? TSMC's like, we don't wanna take the risk. Risk doesn't disappear. It just moves. The risk is right now where you have every single big tech company realizes if we had more compute, we could be making more money. So there's lots of foregone revenue and foregone profits that is the manifestation of the risk that TSMC handed off to them. Risk doesn't disappear. It just gets handed off. And sometimes that risk doesn't manifest in losing money. It manifests in not making money. And there's money not being made right now because what happened was they were very excited about five gs. They did a big wave of like investment, expanding their fabs in around 2020, 2021, '22. And they're like, oh yeah, we're good. Like I said, 2024, Cheshire Bucasal 2022, big thing in tech in 2023. In 2024, their growth rate went down. In 2025, their growth rate went down. In 2026, it's up now. It was very funny because I was writing about this a while ago, and then I think it was like one or two earnings calls ago. Suddenly CC Way, the CEO and chairman is talking about like use cases for AI, the whole earnings call in a way he never had before. This is why the money makers are scared. Usually there's like a bullwhip and they're worried about being at the end of the bullwhip where the demand happens and it works its way down the chain and they're at the end and then they double down. It's already too late. They're wasting all their money. And I think the thing with AI is if it's a bullwhip, it's like the longest bullwhip of all time. There's still so much to be built. And it just took a while for Asia to get the message where these sort of companies are. I think they've by and large gotten it, but them getting the message, it then takes several years for that to actually materialize.
00:42:08 · Patrick O'Shaughnessy
[00:42:08] S0: Do you have a sense for how long you think it will take given the extreme shortage of compute?
00:42:14 · Ben Thompson
[00:42:14] S2: The interesting thing is what this means for Intel and Samsung's sort of logic business. I've been writing about the problem of this dependency on TSMC for years. One of my first articles in 2013 was exhorting Intel. Well, I say you have to build a fab business. You're not gonna be a designer anymore. There's a huge business in manufacturing chips. I thought I was late writing it then. Their stock goes to the moon throughout the twenty tens as they're riding the sort of cloud wave. And it wasn't until 2020 where they finally realized we fell behind. By the way, there's this huge opportunity. We're totally unprepared for it. We don't have a customer service mindset or culture organization, or all the IP building blocks and all these things that TSMC has. And they need a customer. They need customers to help them actually build a real foundry business. So I would write about this as a problem. I write about the China issue. Like you're dependent on a company that is 60 miles offshore of our greatest geopolitical opponent who thinks it's theirs. So like these are big problems. That's where I came to appreciate this insurance issue. For a big tech company to go to Intel and say, Intel, you make our chip. And by the way, the biggest benefactor of this is going to be you because you're gonna learn how to work with a partner. And the biggest pain is going to be us because we're going to have to figure out how to work with you. We could just go to TSMC. They are awesome. They're so great to work with. We know they're going to do a good job. It just never made rational sense for anyone to go work with Intel. That was their fundamental problem. In a unchanging world, TSMC would just win forever. But this is where TSMC and some respects made the same mistake as the mermaid makers made the same mistakes as Iran. If I can continue the analogy. Because they didn't invest the last few years, These shortages are going to be so acute. Big 10 companies that were foregoing so much revenue and so many profits because we don't have enough compute. We will go through the pain of getting Intel up to speed, of getting Samsung's logic up to speed. The scarcity is what ultimately saved Intel. I expect at some point that they're gonna announce some major partner for the first time. It's gonna be a big deal. But ultimately, TSMC brought it on themselves.
00:44:25 · Patrick O'Shaughnessy, Ben Thompson
[00:44:25] S0: It's the cure for high prices is high prices thing. We're gonna route around [00:44:29] S2: There's all these things as like an analyst sitting on the side. You can write these things. And it's one of the things I've sort of learned. No one's gonna pay insurance that they don't need to pay when that insurance expected value is negative. The way to solve the geopolitical problem of dependence on TSMC is to come up with a compute use case that is so massive that everyone is economically incentivized to bring other people up to speed. And then we get the sort of geopolitical insurance for free. [00:44:56] S0: If you think about the, let's say, top 10 or 15 technology companies, which ones do you think have the most interesting setups today for their business?
00:45:05 · Ben Thompson
[00:45:05] S2: The answer is always Amazon. The reason Amazon is so compelling is the extent to which they build for them. They are their first best customer. They provide the scale to get basically anything off the ground, which they then sell to other people. AWS is the most obvious example. AWS, contrary to sort of popular thought, was not spare Amazon capacity. Actually, it took a long time to get amazon.com onto AWS. What it drove was the understanding that we can't be having so many meetings. Like we need to have just compute that you can plug in purely API surface. You don't need to talk to anyone. It's just there. And oh, by the way, if we do that for our internal retail teams, we could do that for anyone. Turns out the retail is so big. We have to start with everyone else. AWS actually started serving external customers before it served internal ones, but now it serves them all. You got other products like say the logistics, where it was the opposite. Right now we're using external providers for logistics, UPS and FedEx and USPS. We need to build this up ourselves. And now they built out themselves. They're offering it to third parties. Other people can use their delivery services. You see this in market after market. They're talking about some of their AI products or their chip products. What's the beauty of the Graviton or the Trainium, particularly the early versions? The early versions were terrible. But if you're on Amazon and you're using some their managed services, like, say, the Redshift database service, they don't tell you what the processor is underneath that. You're just buying a managed service. So they can put all their crappy processors underneath the services they're selling, and that gives them the volume and the capacity to iterate them and get better. And they get to the point where they can actually sell them externally because they were the first best customer for Graviton. Graviton got better. Because they were the first best customer for Tranium. Trainium got better. And now Trainium is obviously running Anthropic and AI products. We'll see if any of them take off. They have call center software. Their call center or their customer experience is going through AI. By the way, it's pretty good.
00:47:02 · Patrick O'Shaughnessy, Ben Thompson
[00:47:02] S0: I haven't tried it. [00:47:03] S2: Moving back to America, I've been buying lots of stuff. Every summer I'd buy lots of stuff in a very brief amount of time. Sometime in like the last year or so, can go on and you're clearly talking to a chatbot, but the chatbot does a great job and it actually does take care of the problem. So you could see that actually starting to work in that regard. But they're building up these AI services for their own business that they're going to make broadly available. And some of them will work. Some of them won't. It's such an elegant approach given they have so many investments in the real world, their core business feels so impervious to AI for the model version of AI. It will benefit from AI, but their moat feels deeper than anyone as far as their core business and their ability to just sort of generate new business lines organically is very compelling. [00:47:50] S0: What about Apple? They've set this whole thing out, it seems.