Invest Like the Best with Patrick O'Shaughnessy
00:57:19 · Ben Thompson, Patrick O'Shaughnessy
Microsoft will help you figure out AI. It will help you figure out in a way where you're not giving away the crown jewels of these companies. We're gonna build this platform, this harness, this sort of middle layer. We're dependable. We're stable. You know us. We have backwards compatibility to the eighties. You can build on us, and then we'll manage all the changing models and what's updating and do all those sorts of things. And does that mean you'll get the absolute best experience? No middleware saws off the sharp edges. You sort of get a lowest common denominator capacity, but if you value in this, the oldest enterprise sales motion, how did Oracle go to market? Oracle went to market in the nineteen days, Larry Allison, with another technology taken from IBM or just IBM didn't want it, relational databases. And they're like, you don't wanna be locked into IBM. Relational database, you could run anywhere. Come with us. The reason this is a joke is because Oracle [00:58:09] S0: Everyone's locks locked in you no end. Right?
00:58:09 · Patrick O'Shaughnessy, Ben Thompson
[00:58:09] S0: Everyone's locks locked in you no end. Right? [00:58:11] S2: But all of enterprise sales is companies whose long term goal is to lock you in, getting you on board by trying to make you scared of being locked into somebody else. All the cloud companies are like, oh, portability, whatever, you can do whatever. They're like, oh, just use our service that only runs in our cloud and now you're locked in. That's Microsoft's playbook. It's a very rational playbook. And I think it makes sense. That's why they have extra money because they're not on the frontier. They are building massive data centers, but they're building data centers for inference. They're not building it for training. And their story about we're investing in time and response to customer demand is more believable in that regard. They're not having to tell a fungibility story where we're building big data centers for training that will be used for inference down the road maybe. [00:58:51] S0: Go back to this notion that it's reckless to not be in the front. [00:58:53] S2: The reason why that's concerning though is at the end of the day, why are we using Microsoft products again? [00:58:59] S0: Because we did before.
00:59:00 · Ben Thompson
[00:59:00] S2: Like to what extent does it actually make sense to have all these artifacts, all these documents, all these email inboxes? Can't AI just do that? There's a real threat here where to Microsoft software business, the whole systems of record thing is funny because one reason why systems of record are so powerful is it's so hard to move them to somewhere else because it's a very tedious, repetitive job. Oh, AI is actually surprisingly good at that. I'm not sure how good the system of record. Microsoft isn't so much systems of record. They do have some of the dynamics business. It's user interface. It's like where you actually interact with the computer. That's the part when you see Codex quad coworker or whatever, it is aimed like an arrow to the heart of what Microsoft has. In the long run, all digital companies are, but Microsoft is very much their strategy is sound. It's also desperate in a existential way and also in a, they might pull it off because they're desperate sort of way. Meta is not threatened immediately, but this is where my bullish view of AI comes in. I think all digital companies are threatened. And Meta is a digital company. They have software. Now, one worry is AI takes up more and more time. Time ultimately is Meta's currency. We saw OpenAI tried the sort of thing. Didn't really take off. Social networks are actually pretty hard. Also cost a lot of money. It's kind of really interesting. So this came up with the creator payment stuff. So YouTube, very famously as paid creators kind of from the beginning. And that's a much bigger drag on the business than people appreciate because YouTube has marginal costs to their content. Now, unlike a Netflix, they don't have to pay that cost upfront. They'll pay it after the fact. So their sharing revenue is a better model than a Netflix model. Netflix is to pay upfront for content and then ideally make more money. YouTube pays along the way. But Facebook or Meta
01:00:54 · Patrick O'Shaughnessy
[01:00:54] S0: There's nothing.
01:00:54 · Ben Thompson
[01:00:54] S2: They pay nothing. Instagram is this unbelievable product that generates all this money for which Facebook pays $0 for content. It's unbelievable. It's funny because you could see a world where for YouTube, AI generated content could theoretically be a positive because the inference cost of generating content could be less than what they're sharing with creators. For Meta, AI generated content to the extent they're the ones generating it is actually a worse margin profile than what they have today because what they have today is free. So they have attention. There's a bullish world where meta is actually very well placed because in a world where we interact with AI all the time, the desire for a human connection becomes greater. And it's sort of like a meta going back to their roots. Meta, one of their biggest mistakes actually, Meta was always a social network company. They killed Snapchat or stopped Snapchat's growth by realizing Snapchat is a great product. Let's layer it onto our network. They brought the network to bear to kill Snapchat. The reason why TikTok was just a blind spot for them is TikTok is classified as a social network, and it's not a social network at all. TikTok is an entertainment product. It doesn't matter who you follow on TikTok. What you see on TikTok is a function of what you watched and you're gonna get more of the same. It's a user generated content network. And the insight from TikTok was the way to get the best content to limit it to your social network is an artificial constraint. We're gonna give you the best content from across the whole network. And the vast majority of content is gonna be crap, but this is like the absolute question before. You don't think about margins. You think about absolute numbers. The absolute amount of great content, even if the margin for great content is infinitesimal, if we have a ton of content, the absolute amount of great content is gonna be very large. I mean, Meta is like, we're a social network. Meta is serving you content from your network of people you know, and TikTok serving you the best content from around the world. That's why they took a huge chunk out of them. Meta had to shift. That's what's happened with Instagram and with Reels is it's not really a social network. It is a entertainment product that pulls from the entire network.
01:02:53 · Ben Thompson
And social networking is like the group chat. It's possible in AI. Actually social network is important again because like we actually want humans. We want to have some sort of connection to them. That'll be interesting to see how that plays out. But the other thing with the models is they're so impactful in advertising. The biggest impact of the models, the biggest monetization right now is probably not in Thropic or OpenAI. It's the incremental gain that is happening for Google and Meta. Most of the stuff is pre LLM, but we're getting to LMs, whether it be generating advertising content. What we want, want verifiable domains. How do you verify if a generated image is good for an ad? Does the ad sell or not? They actually can validate their image creation and their text creation in a way no one else can. And their validation is the ad marketplace. Running a gazillion AB tests on all these different things, see what works, what doesn't. Most ads are a throwaway. It's fine. The vast majority of ads don't convert. They have this massive advantage, this huge liquid market that is a verification machine where the verifiers are humans deciding whether they click on that ad and make a purchase or not, but they're doing it at global scale that can actually have a feedback loop to make their products better. You're also gonna get a world where ad matching is actually still fairly crude. Here's the qualities of the person. Here's the qualities of the ad. And it's like you create an embedding, like a vector calculation and see what numbers match, and then you sort of match and add to the person. What do LMs do? LMs predict. We're going to move to this world where Meta is gonna look at people and say, this person probably wants to see this next. And they're gonna go find that thing and show it to them. The potential upside in terms of showing people better ads that are more relevant to them. They only need to increase a few percentage points for the returns to be billions and billions of dollars. This alone is worth them investing in being on the leading edge in having these amazing models. I think a big problem Meta has is they don't tell this story. It's weird, but Mark Zuckerberg has the same problems Sam Altman does. He doesn't love ads. They have the best ad business in the world. They have an ad business that I think is a societal positive.