Google's AI is winning. Search is paying for it.

It's official, the #1 AI model in the world doesn't belong to Anthropic, it doesn't belong to OpenAI. It belongs to Google. They released Gemini carbon yesterday, and according to the benchmarks, it's the number 1 against any other model that exists out there. Now this is very interesting because Google for the last 6 to 9 months have been trailing pretty aggressively. They haven't been able to put out a capable model and finally, it looks like they have. But that's also the exact reason why their stock.Is plummeting. It's down 12% at the time of recording because AI has started to cannibalize their main business search. Google makes all of their money through advertising revenue via search, and now everyone is choosing not to go to Google. They're choosing to go to AI models like Cha GBT, Cloud, and Gemini. And we were able to get into the numbers this week. Google reported their earnings as well as a host of other companies which reported record cloud revenue, so selling all their compute to AI companies. And despite this, it was still foreseen asA bad thing by investors. And so we're in this weird conundrum where companies like Google, Microsoft, Meta, and all the hyperscales are investing hundreds of billions of dollars in a very capex intensive industry such as AI hoping to generate a lot of revenue, proving that they're generating a lot of revenue, and the market still thinks it's quite bearish. It's this weird conundrum. Microsoft is thinking about laying off 60,000 people. I'll get into that, and a host of other companies, including stock.Markets that don't even exist in the West, the Chinese stock market is down 40%. So the number one question is, have we reached the breaking point? The markets, despite all the good news, despite all the good numbers, despite all the traction, still thinks it's bearish. Anthropic released their S1 this week, and the numbers were a surprise to many people. They're making a lot of revenue at a very lean rate. But the market doesn't agree, there's something fishy going on. And so I want to get into all of this on today's roundup episode brought to you by Qualong.So let's start with the model. The new Gemini model is called Carbon, and they've released it 2 weeks since their last model, which is, again, unbeknownst to Google, who have taken so long to release even one frontier capable model. Now they seem to be churning out every other week. So what's this model even good at? Well, when it comes to general intelligence, it scores 59 on the Artificial intelligence index, which is like the best measure to kind of see.How it performs against all the other models, which rates it as number 1 currently. Now, what it's also good at is long horizon tasks and cybersecurity. This is a smart strategic move from Google because of all the cybersecurity incidents that we've seen. Just over the last week, we've had something called Black October, which is the worst 48 hours in internet history, where basically a bunch or slew of hacks caused by AI.Agents, and we don't know who's behind it, has been orchestrated across the entire world, across various different platforms and the underpinnings of the internet itself. Just yesterday we recorded another episode where Anthropic apparently got hacked and people got into their slack and was able to expose a bunch of different secrets. The point being is we've reached a period where AI models, whether they're old or new, are capable of.Breaking into very secure and reliable software or things that we thought were secure and reliable. And so for Gemini or for Google to come out with a model that is the best at it, currently ranked is a very smart move. Now, of course, you look at cybersecurity, you look at general intelligence, there's one other bucket that's super important, coding. How does it measure against the best coding creators in the world, which is anthropic.The answer is it still comes in 2nd. Clouded Opus 5.5 is the best one and will probably remain to be because Google's kind of struggled to put out a good coding model. OK, so that's the model. What does that mean for the actual market reaction to this and also Google's earnings, like how is the actual core business doing? Now at an initial glance, it all looks quite positive. Sundar announces that hey Google.hit for the first time ever, half a trillion dollars in annual revenue run rate, and our cloud business, which is the business of building up compute and providing it to different types of customers that want to use that compute, the most recent business being AI, which has driven up their revenue, has broken a record. They've brought in over $30.7 billion over the last quarter, which is up.103% from last year, and this has been consistent over the last couple of quarters this year to the extent where they're probably gonna hit $120 to $160 billion ARR for this single unit. By any means, if you were just to look at this company in that lens, you would be super bullish about Google. So why then has the stock dumped over 12% in the opening hours of this morning?There are two reasons. The first is their core business search. Its revenue has started to slow down. The expectations were about 17% growth. They only achieved 12%, and the exec team basically attributed this to AI cannibalizing their search business. Now, the reason whyThis is such a big deal is the majority of money to date has been made via search and advertising. That's how Google keeps the majority of its capital afloat. And for the first time ever, search now accounts for less than 50% of Google's revenue. First time in history.And so people are questioning, is this a good thing or a bad thing. Now the optimistic lens of this is, oh no, that's just because Google's compute product is growing so aggressively. The one that I mentioned earlier that's making around $30.7 billion in three months, absolutely impressive.But the negative view of this is that a slowing down of search revenue is a precursor or a sign or signal that AI is eating into the main core product, and I think about a personal experience that I have when I used to use Google quite a bit, every single day to kind of ask my questions and search for information.And recently I've just been relying on my Cloud or chat GPT or Muse or any other AI product that I can get my hands on because I just feel that it's more intuitive, it knows me, and it's gonna give me a much more effective answer. It can effectively outsource all the searching, scrolling, clicking on links, trying to figure out what information I'm looking like in a singular chatbot. And I think that's starting to affect a much wider scope of Google's customers and the wider world. So the big question.Mark that people have is how on earth is Google going to compete with this? Well, Google's answer is pretty clear in my opinion. They're spending the most amount of money, I think it's upwards of $200 billion either this year or committed to next year to build out their TPUs, their own version of GPUs, compute to train and inference their own models and to kind of like sell. So the idea is they're doubling down on the AI business. The idea here is ifThey are able to build frontier models, which their announcement this week of Gemini carbon proves whatever this new world looks like where people choose to go to AI models to send their search queries, Google will still be able to capture. But I think people are questioning whether they'll be able to actually compete competitively against the likes of Anthropic, OpenAI, and maybe even Meta if we want to throw them into the equation. And so that's why I think people are being very cautious here, because Google hasn't yet proven that they're able to do this. This is the.First time that we've had the changing of the guards where search revenue is slowing down whilst another business unit kind of steps up. And I have to say, the skepticism to a point is actually justified. So far, Google's kind of had a history of releasing these models, releasing the benchmarks that show that it's a good model, and then in practice, it's not actually quite that good. And so people are obviously questioning this new Gemini carbon release, thinking, I don't know if I care enough about this, I'll just continue using Cloud or Chachi.Even if it is better, is it materially better? But don't take my word for it. People who have been paying Google hundreds of millions of dollars to date to use their search ad revenue product are claiming that the product quality has gone down recently. Like, let me give you a fun anecdote here. Google basically pioneered advertising on the internet and it has been the primary way to make money on the internet. You have ads pop up on different websites or on content profiles like this. And so the idea here is Google,Is the monopoly in this case. And one important mechanism that they pioneered is called a second auction mechanism. So let's say Company A offers $1 for an ad placement and Company B offers $5. Now typically you would take the highest bidder, but what Google does is they see the highest bidder and they say, OK, I'll give it to you, but I'll only give it to you at 1 cent more than the lowest bidder. So $1 plus 1 cent, 1.One. And so why Google decided to do this is they wanted to incentivize advertisers to actually give their genuine highest bid and not get screwed over because of it. And so this drove the ad business as it is today to the multi, multi-billion dollar business that Google has. So it's a fantastic model, it's worked really well. What's happened recently, however, is they've gone back on this. Instead of just charging a penny more, they charge the full price of the highest bid.And so people or their customers who have been very loyal to them feel quite shortchanged, and on Google's perspective, you could probably think that they're doing this short-term strategy to boost up their numbers artificially in order to show that, hey, listen, the business is good when realistically it's not actually good. If I was in Google's position, I would just own it and say, listen, things are slowing down, which is what their C-suite exec team has admitted to, but we're doubling down on the new version of what this is going to look like.AI, we haven't started to really lean into putting ads into AI and there's a good reason behind it because Chat GBT, although they have ads, isn't really cumbersome. Anthropic doesn't have ads, and they committed to saying that they're never going to have ads. So the idea here is that Google can't take the first step before these other companies that are arguably ahead of them will end up saying, hey, listen, we have a free product, we don't have any ads, you should just use ours. So Google kind of needs to eat.The cost and loss for now to not risk losing existing customers and bringing on newer customers. So there's gonna be a tumultuous period in my opinion, where Google is kind of needing to tread very carefully, figure out what the new ad model looks like, as well as put out frontier models. That is a lot of pressure. Now, one thing that they're absolutely excelling at is selling compute. Most recently, Apple has rumored to be buying more TPUs.From Google, Apple has a huge cash reserve balance to train their own foundational frontier models. They've also got that partnership with Google where I think they're paying Google a billion dollars a year to get access to Gemini, which is an absolute bargain, but that's a story for another day. And the rumor that broke this week is that OpenAI is reportedly going to be buying TPUs from Google. This is a big deal because typically OpenAI has rented TPUs from Google, and now they're buying, they're investing more heavily as a diversification to Nvidia GPUs. The reason why,TPUs are such a big deal, by the way, is that Nvidia has a monopoly on GPUs and Google and their TPUs are the closest thing to eating into that market share, to being a viable competitor. And Anthropic, OpenAI, Meta have all committed to buying versions of these TPUs from Google, so I would expect to see the cloud revenue go up, up and up. I think over 50% of their cloud revenue was from TPU sales.So it's just a very impressive feat. And the last thing I'll say about their AI business is remember Google are kind of like the forefathers of this entire business or industry. They were co-authors on the Transformer paper, they were OGs in building out infrastructure, their version of TPUs which was like a decade-long love child experience which took a lot of time and a lot of risky bets. So they have all the.Talent, or rather they had all the talent. Over the last couple of months, some of their key players, Jeff Dean left to start his own startup because he didn't feel like Google was providing him with the adequate resources or freedom to go do the thing there, which was such a shame. And then Demis Hasibis, the former CEO of DeepMind, stepped down to take on a new role because, well, we don't know what. So Google is in this very weird growing period.Where it's new management, it's a new team, it's a new model, we haven't been able to access the model, we don't know how good it is. The changing of guard with losing their key market share on search to this new AI thing, it, a lot of it is unproven, but I'm not ruling them out because I think Google is a fantastic company and half a trillion dollars ARR is just absolutely insane. So I think the market is unfairly valuing them and we'll see if my prediction is right. Oh yeah, and I missed one other thing.Another reason why the market is tanking is because they missed on their earnings, specifically around EPS earnings per share, and the reason why is not affiliated to any of Google's businesses, it's affiliated to their investments. They are one of the largest stakeholders or were one of the largest stakeholders in SpaceX before it went public. I think they owned around,14% or something ridiculous like that. And obviously when it IPOed stock pumped and then it kind of dumped and it's now reached around the level of I think it's around $135 I'm not quite sure. Uh, but the idea here is they've taken a loss on the balance sheet and they have to report it on their earnings. So that's affected EPS in addition to search now.I want to end the Google News on a more positive note, which is they released another model, and this is arguably my favorite one. Google developed these things called world models. They're basically AI models that are really good at creating physical simulated realities. So imagine you're stepping into a GTA game or some kind of world simulator.And the thing about Google's products, Genie 4, the one that they just released is it is so hyperrealistic. It's almost like you're watching a YouTube video and it's just completely artificially created. I mean, look at some of these demos, you have one of a guy on a motorbike and you can see someone typing in saying I want the other motorbikes to overtake me, I want the.Traffic lights to turn green, and you'll notice that you can hear certain sounds that generates. You'll notice that you can actually speak and interact with other people in the world. There's this great example from Ethan, Ethan Moloch, I believe, who created this game, this Neolithic, Neo-Gothic game. He asked to spawn a beaver that walks up to him and talks to him, and it happens in an.Instant, and the beaver talks, it has a voice, it generates all artificially. Now, what's interesting about this product is the AI model powering it is obviously great at video generation, images, etc. but it's Gemini 3.8 Flash, an old, cheap model is powering this brand new wonderful experience.And the reason why I wanted to mention this is I'm noticing a trend, which is the best or some of my favorite AI products today or this year have been these things that aren't powered by the frontier AI models. I don't quite know what to make of this, but I just wanted to point it out because I think there's been an assumption historically that the best AI models give you the best AI experience, and certainly we've seen that with coding, solving math and whatever that might be.But I think there's a new bucket or sector opening up where the consumer apps don't have to be powered by the best models, they could be powered by open source models or cheaper frontier models that unlock a new experience. Imagine just because it's cheaper doesn't mean it's bad, it means that you can spend more tokens to do crazier things like generate realities in an instant and be able to hear, speak and interact with things.Imagine what this means for applying it to gaming, or applying it to movies, or applying it to simulated realities which you need to train robots or whatever that might be. These are all applicable things and my big bet of the year and the next couple of years is that world models actually end up competing pretty directly with LLMs. We'll see if that pays off. OK, so I wanna get into a few final thoughts on today's episode brought to you by Q Qualcomm.Now next story, Microsoft had their earnings, but that's not the focus point of this week. The focus point is them potentially laying off 60,000 people, which is 25% of their entire workforce, including their AI teams in the name of AI Capex. It should free up around $70 billion that it will spend to buy more Nvidia GPUs, set up new data centers, and fuel the AI beast. This is a questionable move because Microsoft has spoken the benefits.Of AI for so long, I believe Satya has even said that like, listen, this should lead to a transition of roles and jobs, not actually loss of jobs, and we're seeing kind of the opposite thing take place here. Now, I understand why they're doing it. Capex is very expensive and you need to kind of keep the business, especially as large as Microsoft, as lean as you can. They spent around $130 billion this year. I think they're gonna do the same next year, but what's crazy is they're also making that same amount this year. So the question is, are they?Allocating resources in the most effective way, doesn't make sense. The other thing that struck me is, I can't believe they're getting rid of their AI teams. Now, I know that Microsoft has a major financial stake in OpenAI and other such AI providers. I know that they've been very open to allowing more recently, other AI models to live on their platforms. They've become model agnostic, they got into a kind of like a bad fight with Sam Altman and Chad GBT and they were like, OK, cool, we'll become agnostic. But they were also.Working on their own frontier models. I believe the models are called MAI and a host of other things. One thing has been pretty clear, they have not been able to catch up. Google was behind and now they've been able to catch up. Microsoft has failed pretty much the entire time. I don't know whether this is because they have a lack of high quality researchers, they're indeed hard to come by, but this kind of seems like they're hanging up the towel. They're like, OK, cool, we'll just become one of the number one neo cloud providers, and if you look.Their recent earnings, they've absolutely crushed it, specifically on Microsoft Azure. Google Cloud still beat them, but you know, it's still a good business, but it seems to be like they don't care about building their own AI models right now, and it's just interesting to see and kind of worrisome to see about people potentially losing their jobs. Now the buyout presumably is around 6 months plus more, and so they're not doing this for cheap. It's obviously worth the cost to them, especially as the cycle becomes more Capex heavy and intensive and.That's the other thing, Google is committing around 200+ billion dollars for next year, Microsoft is doing a similar figure, Meta's committing the same figure. And so we're seeing this growing trend of we're just doubling down on this AI thing, especially when companies' cash balance sheets are going into the red.So we'll see how all of this plays out. And the final story I'll give you is that the earnings aren't just a Western thing this week, it, they're crushing it in Asia as well. SK Heinx, which is the number one AI memory provider, slam dunked it. Their revenue is up, I think 328% year over year, and this is important because for a number of quarters before this, people have been calling the end of the AI memory trade.And what is clear is that not only is that false, but two, memory is sold out through 2029, which means three years ahead of where we are today, or 2.5 years, people are still buying up memory. The craziest part about this is around 50% of those contracts are long-term contracts, which means that all these customers who are committing to the memory years ahead have paid upfront likely 50% of the cost. So you have.This memory provider, which is very well financialized, that are building even more higher capacity as quick as they can, but they can't sufficiently provide to the demand that they're seeing. Now, you might think that this is bullish for Asia in general. It's not. China's stock market has tanked 40% since its June highs, and people are thinking that this might be a precursor to the AI trade and AI bubble in the US. Certainly we've hadFigures like Ken Griffin, Elon Musk called the top of the AI bubble. I think yesterday I saw, uh, Snowflake's co-founder and CEO say the same thing. So again, we're reaching breaking point, that's the whole theme of this entire episode and this entire week is, despite all the good news, despite all the record earnings, despite all the crazy numbers, the S1 findings, yada yada yada, people still think that this might be a bearish turning point for AI because.It feels almost too good, and I can't say that I disagree with them too much. I don't have the answers right now. All I can do is look at the fundamentals. Now, I'm very optimistic on this show and I will continue to do so. I think a lot of easy takes to have on AI is just assume the bear case and that gets more views. I don't care for that at all. I truly feel in my heart of hearts that AI is gonna revolutionize and change the world. It might just take more.And right now we are overcapitalized. We might be taking on a little too much leverage and honestly, I'll dedicate an episode to this next week, but I still think ultimately we're directionally ending up in a good place and maybe we see some healthy consolidation in the meantime. Thank you so much for listening to today's episode. If you're listening to this on YouTube and you aren't subscribed, please subscribe, turn on notifications, do the same thing on Spotify, Apple or wherever you're listening to it. Our socials are also linked below. And yeah, that's today's episode brought to you by Qualcomm. Thank you.