Invest Like the Best with Patrick O'Shaughnessy
01:09:36 · Ben Thompson
[01:09:36] S2: stuffs, etcetera. But implicit in that, why do they get a lower cost of capital? They get a lower cost of capital because NVIDIA assumed risk. This is my point before. Risk never disappears. It just appears somewhere else. Taking on risk has a price. There's a world where AI takes off, it never stops, and everything is fine. And NVIDIA captured all the upside of their risk. But there's also a world where, say, that this is Neo Cloud that backed up, A ton of compute comes to market. The hyperscalers have plenty of compute. They don't have enough compute. NVIDIA is paying for a compute that no one wants. They just lost a bunch of money. If you think about it, there's an expected value of that investment. That expected value, it's not zero. It's not 100%. It's somewhere in the middle, but that is a diminution of Nvidia's profitability. If you actually look at their business holistically, what that is, is a price cut. Now the price cut didn't show up in margins. It didn't show up in what they're offering, but a lot of what Nvidia is doing is how can we maintain our margins even if the wide view sort of discounted cash flow, expected value, holistic view of our company. People do discount cash flows, but are you actually considering all these pieces? The reality is is that moving stuff off the balance sheet by and large works. But they're doing all these deals to maintain what feels somewhat unnatural. We have seen price cuts. They're just manifesting in these very bizarre sort of ways. Now in the long run, I think the challenge is their ultimate competitors are the hyperscalers, particularly Google and Amazon. So Google and Amazon aren't just building their own chips, but they're also looking to sell those chips externally. Google already made a deal to sell, like, 20% of their TPUs, Genthropic. On the last earnings call, Andy Jassy practically confirmed that they'll be selling Tranium threes or maybe Tranium four, or they'll be selling Tranium chips sort of eventually externally, which makes sense. That gives them a long term buy into these companies. There's a huge amount of R and D that goes into developing chips. They get more leverage on their spend. It all makes sense. And by the way, they're not selling their chips on differentiation. They're selling their chips as commodities. NVIDIA is the one selling differentiation.
01:11:36 · Ben Thompson
People aren't going to Amazon to use Tranium. So they're not cannibalizing the attractiveness of their cloud by selling Tranium outside. So they're NVIDIA's biggest problem. Because what's the number one advantage that the hyperscalers have? Scale. Lower cost of capital. It's a capital fight. They have a lower cost of capital than the Neo Clouds do. The Neo Clouds are... They'll buy NVIDIA left, right, and center. And by the way, it also makes total sense that why SpaceX satellite, Elon's out there, we will always buy NVIDIA because they're the best. No. You'll buy NVIDIA because they're the most fungible. NVIDIA is true. It is the most fungible. CUDA's moat is dramatically diminished because the models don't care what they run on, and that's what actually matters was built on top of the models. But it still matters. It's still something of a moat. It's if you wanna play the game SpaceXi is doing, where we're gonna build a lot and rent it out, but reserve the right to pull it back, of course, you're gonna be on NVIDIA because the easiest way to rent it out is to be on NVIDIA. You saw this very early, by the way. You go back to 2024, 2023. NVIDIA starts talking about all these sovereign clouds. They start talking about... They tried to call these Neotron models. They had this thing in 2024. I remember. It was the first one where it was like the rock star GTC at San Jose and like the huge Colosseum and just one comes out. It was a very boring GTC. The old ones used to be NVIDIA demonstrating like 50 gazillion things as they're throwing stuff at the wall. They knew they had something with GPUs, and they're trying to like find these
01:13:00 · Patrick O'Shaughnessy
[01:13:00] S0: out of them.
01:13:01 · Ben Thompson
[01:13:01] S2: Once LM showed up, it's like, oh, we have the use case. But they were coming up with all these enterprise offerings. I can't remember what they were called, but they were like these modules basically that, of course, they were free, but they only ran on NVIDIA. And you could see what they were doing is they were trying to lock people in. Intel is a good example here. AMD cleaned them out in hyperscaler sales. Because the hyperscalers are putting the effort to get stuff working on AMD versus Intel. There are still small differences, even though they're x86, because they're buying at such scale. The investment to do it is worth it to get a better chip or lower price or whatever it might be. The part of Intel's business that never floundered was selling to government and selling to enterprises. They don't have the resources of a hyperscaler. They're not buying at that scale. They're just gonna keep buying what they had before. That's why NVIDIA talks about selling to sovereign clouds. That's why they talk about selling to enterprises because they wanna get in these markets where they're not gonna be balancing this chip versus that chip. The hyperscalers have always been the threat to NVIDIA for that reason. They're actually bigger. So you have this issue where the hyperscalers are the threat. The hyperscalers have a better cost of capital than the other companies NVIDIA wants to buy them. That's how you get this deal this week. I see this deal as a response. That's why it goes with the Google deal. Google can just issue equity. The shareholders don't love it, but their monetization capacity is much higher than NVIDIA or NVIDIA's customers are. I think what NVIDIA is hoping for, maybe they wouldn't say this in so many words, but if we get to a world where we actually run out of power, that's probably good for NVIDIA. Because in a world where we're totally constrained on power. Everyone wants
01:14:38 · Patrick O'Shaughnessy, Ben Thompson
[01:14:38] S0: the best. [01:14:38] S2: We have to get the best efficiency. Yeah. The best token efficiency. And I think NVIDIA is still the most token efficient. So that is a good world for them. Probably the biggest problem for NVIDIA over the last couple of years is I think The US has actually brought a lot more power online than expected. It surprised me. Whether it be what Elon did sort of behind the meter, which has been replicated or West Texas and natural gas, but even like restarting nuclear plants. [01:15:02] S0: You love how The US responds to these things.
01:15:04 · Ben Thompson
[01:15:04] S2: It's awesome. It's actually one of the biggest encouraging signals about The US is I was writing early on, assume this is a bubble. You want there to be a long term payout. The.com, we got fiber in the ground. And by the way, Google has played this game before. Google built its business by buying up dark fiber. They had the killer search engine, but so much of the power they do is because they bought up all this dark fiber that was basically free after the .com era. Our core Internet still runs on WorldCom fiber. That was a lasting benefit. The railroads, BNSF is throwing off money that's going to Google from Northern Pacific and Jay Cook selling bonds to retail investors. You want a bubble that produces something that lasts. And very long, it's like, what's gonna last from AI? The GPUs don't last that long. Data centers. Okay. Fine. But what is it gonna be? Power. It has to be power. If we're in a world where this all blows up and we have way too much power, that is an amazing world to be. We've always been energy constrained. Energy undergirds everything. What would it be like to live in a world of energy abundance? It's hard to even imagine because our minds are so constrained by the fact we've actually always been in energy scarcity. I think we've done an unbelievable job. Power for sure is a constraint. It's going to be a constraint, but I think it has taken longer to become a constraint than anyone expected. And I wouldn't be surprised if that includes Jensen Huang. I think he thought insufficient power was going to be NVIDIA's moat sooner than it happened. It turns out that the longer we have enough power, the more time Amazon has to make Tranium better. The more time Google has to make TPUs competitive from an efficiency standpoint. And if we get in a world where...