00:07:53 · John Coogan
And so that is the is the coming challenge, but, you know, they've been doing this for twenty plus years. And so that is the path that they're on. The last thing, Google is spending $200,000,000,000 on AI infrastructure this year. 2026 CapEx guidance, dollars 195 to $2.00 $5,000,000,000 right in that $200,000,000,000 range. They are also signaling that they will expect that they expect a significant increase in 2027. Q3 spending alone was 51.4. So they're already on their $200,000,000,000 run rate, and next quarter should be higher, following quarter should be higher, etcetera. So let's go. And then the last thing is that Gemini crossed a billion MAU, monthly active users. The first party AI model, the API is doing 32,000,000,000 tokens a minute. Hard to say what that means in the modern era because there's so many models that are more token efficient and really just getting stuff done, getting answers, satisfying users is what matters. But it's up 40% quarter on quarter. Clearly, they're serving more models. More people are using them. And then, of course, they announced Gemini four Carbon, the newest frontier model, the day or before earnings, two days before earnings, sort of trying to reassure investors that they are still on the frontier, that they're... That the benchmarks look good and that the adoption will come and more importantly that they are vertically integrated. So even if they don't show up on the ramp data and the open router data, just having the frontier model internally is super valuable because it will drive YouTube, it'll drive AI search, it'll drive internal workloads. All... There'll all be all these things where they aren't going outside for costs and they will be vertically integrated.
00:09:40 · Jordi Hays, John Coogan
[00:09:40] S2: Yeah. Seems like it's gonna it's gonna be incredibly strategic to be able to get Frontier Intelligence at Yeah. 90% low, you know Exactly. Lower than market prices. [00:09:49] S0: Exactly. Exactly. And then there's a whole bunch of other interesting things. They can fine tune their own models. They can optimize them for their infrastructure. What do you... The model that works in Google search, it does feel different than a Claude model or a GPT model because it needs to be really fast because people expect that from the AI search results. And then on YouTube, you expect something different. You expect something that can piece... Can crawl over all the transcripts of YouTube with no, you know, robot block up, no captchas and whatnot if you're asking for extra context. So lots of lots of optimism there, but rough day in the market at least. [00:10:30] S2: And a little bit rough for Berkshire, who their sort of blended price per share of their recent $10,000,000,000 investment is around $350 per share. They did some part into Class A, some into Class C, down pretty dramatically from that point. Alphabet now sitting at $280 a share in just around six months. [00:10:58] S0: Got it. Well... [00:11:00] S2: And, yeah. It's like the OG Google bear case is like rearing its head again. Yeah. [00:11:05] S0: Right? Yeah. That's right. [00:11:06] S2: And, you know, they've been... It became sort of There was a viral thread. I'll try to find it. What are [00:11:15] S0: talking about from levels.io maybe?
00:11:15 · John Coogan, Jordi Hays
[00:11:15] S0: talking about from levels.io maybe? [00:11:19] S2: This was... Basically this guy, July 23, Max Anderson says, As someone who has personally spent 500 ks a month on Google Ads for years, I can tell you with certainty this revenue growth in search is artificial and extremely unhealthy for Google's business long term. He goes on to make a bunch of points, which have been refuted in different ways, but Sattrini brought it up yesterday. The souffinator at the time pushed back on a lot of this, but either way, it just felt like search was already perfectly optimized. I think I was bringing this up probably over a year ago. Have you actually paid attention to a Google search recently? Where it's like, for certain queries, it's hard to find an organic result, right? Where it's like ads at the top, first three results are ads, then you have maybe one or two organic results and then a bunch more ads.
00:12:12 · John Coogan, Jordi Hays
[00:12:12] S0: I mean, have have you... Do you understand the structure of, like, finding a mattress on Google? It got so crazy that, first off, there would be Google Shopping callouts that are also paid for. There would be a, Google Shopping widget. Then there would be three or four sponsored results. So the links would be, you know, specifically this brand, specifically this brand, just directly brand keyword advertising. But then you'd go down and you get to mattress review sites that were like... And they'd all been purchased by the various mattress companies. So, like, there was actually no where to to end up. There was nothing that you could go to. There was no, like, Wikipedia. And and, yeah, it's been it's been crazy, but but this stuff's just sticky. People are just in their Yeah. Flows still. So that's why you only see, like, slight deceleration. It's not like falling off a cliff or anything like that. [00:13:02] S2: Yeah. Anyway. One other interesting detail. So Microsoft had earnings as well. Yeah. CapEx came in at $38,370,000,000 Estimate was $45,900,000,000 So Google was $51,360,000,000 Estimate was 55,700,000,000 This wasn't because they didn't want to spend the money, it was because of delays. So they're like, they're trying to spend Powered shell delays. Yes. Just a variety of
00:13:26 · John Coogan
[00:13:26] S0: different Yeah. This whole rumor of like, there's a whole bunch of GPUs that are waiting to get racked because stuff is slow and there's permitting and you're bumping up against the real world, the physical world. It doesn't really matter where demand is And everyone who's selling tokens is higher is pretty happy to be like, yeah, we wouldn't hate if our margins were good this quarter even though we were delayed on supply. Being supply constrained isn't the end of the world for most people. Well, on the AI build out and strategy specifically, I'm new tie, new, new, I'm just new at AI, I guess, has a breakdown of some rumors saying that the internal model's reasoning jump comes from more aggressive RL, including training in games and simulated environments. Simulation hypothesis is going strong. Strategies learned their transfer to research mathematics and other fields, and obvious Stokes and ETH show the scale of the improvement in Anthropics in a similar direction with its internal quad model. Musk's digital optimist hiring explicitly calls for manufacturing measuring transfer from games into general computer use. And so Ashley Vance over at Core Memory is saying that, Does this mean we're just in the midst of Demis Hassabis' original vision of how to get to AGI? After a kind of brief brute force period, that games are the path to AGI. And it is interesting that that Google has stayed the course with Genie when a lot of the other labs... Mean, Anthropic has famously been like, we're only focused on coding. We're not even doing images. OpenAI has done a bunch of, like, quote unquote side quests. Some of the side quests were too compute intensive, like Sora. Some of them, like, computer use. Remember operator? Do you remember operator? Like, it was the original, like, go and computer... Use a computer and a virtual computer to go and click on buttons on a website. And it was pretty janky at the time, but now it's basically solved. And that was actually a good investment on the side. And it'll be interesting to see which way Genie goes. Like, it could just be, oh, it's only fun and it's really expensive, or it could be like, hey, this is a really, really great next leg up for training potential, it's a really valuable asset for them to have internally. I think that's what people are are debating right now. Anyway, where do you wanna go next?
00:15:50 · John Coogan, Jordi Hays
Let me tell you about the New York Stock Exchange. Wanna change the world? Raise capital at the New York Stock Exchange. Wanna stay on [00:15:58] S2: Google News or move in video world. Yes. Should we start with this, this Times feed? [00:16:04] S0: This is in the New York Times. Although they had posted something similar a year ago, right? Yeah. Take us through this and then we'll reflect.