Are LLMs Hitting A Wall, Microsoft & Alphabet Save The Market, TikTok 'Ban'

0:00Are LLMs hitting a wall? Microsoft and Alphabet might have just saved the market and the TikTok ban is on. We'll talk about all that and more on a special Friday edition of Big Technology Podcast coming up right after this.

0:42Welcome to Big Technology Podcast Friday edition, where we break down the news in our traditional cool-headed and nuanced format. We have a special guest for you here today to help talk through the big tech earnings, along with the latest AI news and, of course, the TikTok band. Deirdre Bosa is here from CNBC, one of my favorite people to watch on CNBC and to speak with. It's great to have you here, Deirdre. Thanks so much for coming. Well, what a coincidence. This is one of my favorite podcasts to listen to. So I'm so glad we could collaborate. I'm very excited to be on. And what a week. There was so much to get to.

1:14I know. It's crazy. I mean, it's so funny because you kind of think you know the story going into a big earnings week. And then everything gets flipped on its head. Companies miss and their stock goes up. Companies beat and their stock goes down. And then you end up having real check-ins with companies like Alphabet and Microsoft in terms of like where this market is going and, you know, some narrative reversal there as well. And for us, sort of the whole markets hinge on a few of these companies, right? Because they just make up more and more of the broader market. So whenever it comes along, we like to say at CNBC that it's our Super Bowl, but we have a Super Bowl every quarter.

1:52Exactly. I know just a series of Super Bowls. And that's kind of what makes this news fun to watch. And one of the things that I've been watching over the past week is just the amount of resources that these large language models consume. And I'm curious what you think about this. So last week, when I was speaking with the head of generative AI at Meta Ahmed Aldala, he said that they use 10 times the amount of data and 100 times the amount of compute to go from Lama 2 to Lama 3, right? And effectively, there was a story in the New York Times talking about how they needed to buy Simon & Schuster or something on that magnitude to get enough data to to be able to keep pace with chat GPT.

2:28Then you have Mark Zuckerberg on the Dworkish Patel podcast talking about effectively like the biggest constraint is gonna be energy and you might need a nuclear level plant in order to keep running these models. And so the idea of what these things have always been, you throw more data, you throw more energy and you throw more compute at them and they'll continue to improve. And I'm curious from like the people that you're speak with in your view of the industry, whether you think that, you know we might be close to hitting a wall or do you think that this is something to be worth paying attention to because the principles are speaking out loud about it yeah and so there's compute power on one hand which requires enormous amounts of power and that requires enormous amounts of capital um on cnbc this week we had dario amadea from anthropic and he was saying that he was talking about the idea that these large language models would be commoditized and He said the costs are only going up to your point.

3:24By the end of this year, to create a large or to build a large language model, it's going to cost something like a billion dollars. But he said by next year, 2025, 2026, it could cost up to five to ten billion dollars. And that's really going to narrow the field. I thought it was interesting because the same day that he said that there was this article out from the Nikkei in Japan saying that SoftBank, Masa-san, was looking to develop a Japanese large language model. and they had earmarked up to nearly a billion dollars. And you just think that's not going to be enough. And even if they do get that money together, they need the compute power.

3:58They need the energy to be able to do so. But Alex, isn't there another camp that says that the chips, everything about this is going to become more efficient so the cost will actually go down? Yeah, definitely. And I wrote about that in Big Technology this week, basically saying that if we keep going with the way that we're going, we're going to hit a wall or the tech industry really will hit a wall with LLMs. like they're basically if you think about it out of data almost out of compute and out of uh and running out of energy here and so it is going to take these optimizations in terms of models and chips uh and even potentially more energy efficient solutions to make it work and they're going to do that all the while they're trying to get more and more of the natural um of the ingredients here right so we're thinking about like you know there's going to be synthetic data that they're going to add so they'll have more data and then you know maybe someone does go build that nuclear your power plant.

4:50Like seriously, like Zuckerberg's like, it's just a matter of time until it happens, which is crazy to think about. And then you have Sam Altman running around and saying, let's get$7 trillion for chips. And so, you know, what you might end up having right now is this kind of dual action where you have effectively companies that are finding ways to build these things more efficiently. We think that's happening already off across the board. I speak with an Amazon executive yesterday who said that there's some researchers in Stanford who are learning to train these things faster with the same amount of data and 10 times less cost for inference and it's like all right so that's happening and then on the other hand the other side of it is we these executives see how important it is to have these ingredients and they're doing whatever they can to expand and you're seeing that show up a lot in like the cloud infrastructure numbers and capex numbers from big tech we saw this week is they're just spending so much more in this race um one of the best ways i heard this described is that like a typical or traditional Google search takes a certain amount of compute power.

5:52And then for a chatbot search, a chat GPT or Gemini or whatever search takes 10 times the amount of compute power, even though it may just be a simple question, like two plus two equals four, they're still running the whole gambit to get to that if you're asking it through a chatbot versus a traditional Google search. So like that efficiency, it seems crazy to me that that's not going to be figured out soon. Exactly. And I mean, you know, we can get right into the earnings now because it sort of was a theme for everybody, but especially for Meta. And with Meta, it was like they are doing like they increase their ad revenue tremendously.

6:30And you think about these companies going off like the bases, right? Like you're in the$30 billion a quarter range and you're still growing at like, you know, 15, 20, 25%. That is unbelievable. But what happens in this earnings call is Mark Zuckerberg goes ahead and says, listen, we're in this era where we're going to have to spend a lot on AI. It's not going to get more efficient anytime soon. And this has happened to us in a number of areas like reels and stories and the move to mobile. And you're going to want to hang along with us for the ride, but it's not going to happen overnight. And then all of a sudden the stock goes down like 18 % after hours and everyone's just like, all right, meta's done and i guess i don't understand why why investors sort of bail on a company like this like it seems like very sensible is it just the moment that we're living in right now that's just like there's so much pressure because of the higher interest rates or is there something else going on there we talked about this earlier this week in the context of a piece that we're doing over at cbc which is like the roller coaster ride that is mark zuckerberg right facebook and meta and And coming off the year of efficiency, which he really pioneered for the rest of big tech, he made it okay to do lots of layoffs, to get fit, kind of coming on the call and saying, okay, that year of efficiency was great.

7:50And we're going to keep that kind of discipline in terms of our workforce and certain costs. But when it comes to generative AI, we are all in and we're going to be spending tens of billions of dollars. I kind of thought he was just more candid than the other studios. When I say others, I mean, Google and Microsoft, they didn't give numbers and they try to say, listen, we're monetizing generative AI, but look through it. We don't actually know. Sure. Maybe they're monetizing it around the edges, but they're not breaking it out. They're not telling us exactly how much we're just sort of trying to look through the cloud numbers and say, okay, maybe this is coming from generative AI.

8:25Where Zuckerberg was a lot more, I think, upfront and transparent saying, listen, the payoff for this is some years out. And yes, it's a different model, right? He's basically giving it away for free right now. He doesn't have the cloud platforms that Microsoft and Google have. His company is not a hyperscaler. So yes, it's further out. But I don't actually think the story is all that much different than what we heard from Amazon and Microsoft, which had huge reactions. Part of it, though, I will say is that Facebook has a lot smaller revenue base, right? So as a percentage of their revenue and percentage of earnings, they are spending like kind of an unbelievable amount.

9:05And he's also spending on the metaverse. Exactly. Got to keep those headsets going. And yeah, I mean, the revenue is incredible. So I'm pulling up the actual numbers. It was the jump 27 % to 36 billion on the quarter up from 28 billion the year before. And this is amazing because, you know, analysts have talked to me about, hey, look, like there's only so much more room to grow in digital because you're just taking all the money from, you know, these other formats and you're going to eventually just hit a wall because they'll still remain relevant. You don't have that much more time spent. And for them to increase 27 % is crazy.

9:38I mean, maybe it's taking money from the rest of the open web. But you're right. You pair that money with sort of Zuckerberg's message of it's the year of less efficiency, right? He's saying by building the leading AI, we will also be a larger undertaking than the other experiences we've added to our apps. And it's going to take several years for this to play out. So it is very interesting. It's I'm curious, like from an investor standpoint, do you just kind of like run away from from the stock when you see that there's going to be increased spending? Or is there a case like, you know, I was seeing in my feed basically right after the market punished Meta so bad that everyone's like, all right, this is the buying opportunity.

10:19I mean, I know you're usually the one asking those questions, but I'm curious what you think. It's a great question because the morning after Meta reported was also, I believe, the morning we got GDP. So if you were watching CNBC, you would hear on our air all morning fears of stagflation. GDP came in lower than expected. Stagflation is obviously an investor's worst case scenario. We haven't seen it in decades and it's very scary for the market. So there's this idea that you do flee for safety. You don't flee to a company or you get out of a company whose earnings are way out in the future. Right.

10:57And I make the argument often on air that the mega caps are safety. They've got pristine balance sheets, huge moats, no like very little debt. But when you just heard Zuckerberg say, we're investing billions and billions of dollars and the payoff isn't till many, many years in the future. You just look at your discounted cash flow model. If you're an investor and you say, I'm not going to buy this right now. I'm going to go to something that gives me earnings right now, especially when interest rates are where they are. And having cash in the hand is actually, you know, a pretty good proposition right now or something like gold, which has been rising and rising this year.

11:31So I love that you asked that question. You always have to put tech, even the biggest tech companies in the context of what's going on in the macro environment. But I wouldn't be surprised if you see what happened last year when interest rates were rising and when they were higher than they've historically been over the last decade. Investors come back into tech because of the reasons that I first listed. They're actually kind of like these giant consumer staple companies because of their cash flows and their cash piles. And on that point, too, it was so interesting because it was last quarter that Zuckerberg said Meta was going to issue its first ever dividend.

12:09And the stock went crazy, went to the races. So it's almost like we forgot that, spending billions and millions of dollars. But it has so much cash flow that it can still return cash investors, which is very not typical from a techie growth company. And, of course, that could lead us into Alphabet's big announcement yesterday. Yeah, yeah. Yeah. So, I mean, Alphabet announced this dividend. Before we get into it, let's just touch on this point that we spoke about earlier in the week that just like it is really a roller coaster for Meta. I mean, like last week you had Zuckerberg and the fur coats like releasing these AI models on top of the world.

12:45And this week he's like, we're going to spend on that AI. And everyone's like, what the? You brought up this point. He's one of the last of a founder CEO generation. He's one of the last guys standing along with Jensen Huang, who, I don't know, has the boldness or the confidence to or I don't know what it is, just this founder quality to have a lot of conviction in what he's doing and not necessarily do everything for Wall Street. Like you have some of these operators who have who are more I don't I don't know what the word is like. Zuck's done a really good job in speaking to Wall Street, like better than some of the other operator CEOs that we have.

13:27Um, but he also, this just proves that he's not afraid to, you know, upset them too. If he thinks something requires a lot of capital and, um, a lot of risk. Yeah. And I definitely want to get to Google, but I, I kind of think let's, I'm dying to talk about Microsoft because it's kind of the other side of this, uh, you know, whether AI is going to sort of prove itself out. So we have this no business model kind of, uh, I mean, maybe not no, but unproven business model, give everything away from for free. Although I guess they are going to license Llama out through the cloud services, but that doesn't seem like it's going to be meaningful revenue for Meta.

14:03And then you have Microsoft, which is just like, yep, shipping it out, longest partnership with OpenAI, and it's currently in market. And so, like, we're starting to see, like, how valuable this stuff is, actually. And I can't tell whether the numbers that Microsoft announced yesterday are big or small and would love to get your perspective on it. So, first of all, they beat on earnings and revenue and their stock is up. So between them and Alphabet, they seem to be saving the stock market. But this is from the report that Azure rose 31 % during the quarter. And Microsoft said seven percentage points of the Azure growth came from its AI services, up from six percentage points the previous quarter.

14:41And, you know, if you're telling me AI is a revolution, not you, but them, you know, they're telling us that AI is a revolution. Like, all right, single digit percentage point on cloud growth is nice. Am I just being impatient or is there some real concern that this is not going to translate? I think the story with Microsoft has always been that they're well placed, right? And I think that Satya Nadella has done such a good job just saying like, listen, we're out front in this. We're going to make Google dance. We're going to do all of these things, right? So stick with us. We're the play. But I completely agree with you that revenue contribution is so unclear still.

15:16I think maybe investors realize and accept that. But 7 % of Azure growth, I don't even know what that means. That could be like, we need to know what exactly that looks like, right? That could just be like some services that they renamed AI or some existing services. And then when it came to co-pilot, like no details either, just kind of like some mushy talk around how important it is and no hard numbers on how well it's selling. So I agree with you. I mean, they have all the right partnerships. They're saying all the right things. Just don't look too hard at the numbers. Yeah, right. And this is one of those interesting things about this AI moment is that it's progressing so fast.

15:58Like the innovation is like the fact that I can like live type and meta AI and see an image appear before my eyes is staggering or upload documents like many meetings to Claude and ask it to synthesize them is incredible. But we also just like don't really know exactly what the revenue picture is going to look like. you were going to say something about Claude. Well, because I was kind of like pointing at you because I saw your tweet, which made me so curious. You said you were happy to pay for Claude,$20 a month. It's the only one that I don't use. And I switch a lot between Gemini, ChatGPT4 and Perplexity.

16:34I haven't used Claude and your tweet made me really curious to try it. Yeah. So now I'm going to vote Claude head. I used to be a Bing boy. You know, you just undermined your whole argument. I know. Ranjan, who's usually with us on Fridays, he's a Gemini guy, so we'll let him defend himself next week. But he actually is the one that turned me on to Claude. And I think one of the cool things about Claude is just uploading documents. And one of the things that I've been doing recently is uploading full podcast transcripts. And of course, like initially, I was like, I'm just going to speak with Claude about like what happened in the show and like which questions I missed.

17:12And I'm also, I love saying rate this conversation on a variety of metrics. And it really does. It does on chemistry and depth and breadth of the conversation. Has it ever offended you? No, it's actually been pretty good on that front. There's even been points where I've been like, you need to rate one aspect of the conversation, one of 10. This is not an option. And it's like, well, yeah, actually the weakest part of the conversation was this. But I've been using it for recently is actually I go in and I say, listen, I have to make some video clips from my show. Give me some timestamps of things that you think might be compelling.

17:48And it does an incredible job. This used to be something that I wouldn't do because it would just be so time intensive that there was no justification for like sitting in the video and finding the perfect timestamps. And Claude is actually, I'm like, all right, those are good suggestions. Let's clip those. That's really fascinating. I tried to do that yesterday. Actually, we do this Tech Check weekly piece which is sort of a deeper dive on the week's biggest topic and i put it into chat gpt4 and i said give me you know like a condensed tiktok version of this and it was so bad i just was like i quit right here now but i think that's an interesting way of doing it look at the time codes they can't do it all for you they can never do everything for you at this point but um i'll have to try it out yeah i mean so they have this like uh paid version that's gets you their most powerful model, which is Claude Opus.

18:35And I used to like, basically, the reason why I sent that tweet was, I kept like, logging in, signing up and then immediately canceling the auto renew, because I'm like, Yeah, I'm only going to need this. I'm trying it out. And this was the week where I was just like, I'm not even canceling the auto renew. I need this. So it's like now definitely part of the mix. And is it going to cost me$240 a year? Or maybe a little bit more with the fees? Yes. Is it worth it? You have multiples. So, so, so that's like, oh, sorry, go ahead. On a slightly different note, I was telling, um, I produced a lower bachelor yesterday how, and this maybe gets us into Google earnings.

19:12Um, sorry if I'm jumping the gun here, but my method of working now is to have, and I guess my, I'm a GPT four person, your Claude and Ron John is Gemini. Is that right? Yeah. So I think, yeah, we'll put you squarely in second place here. Sure. I'll take it. But I was telling her how I always have that window open now and I truly, truly use a traditional Google search. I use Google news a lot too in the past, but I use it way less than I ever have. And it truly is kind of my co-pilot to use that word that is so, you know, buzzy and hot right now. But I use it, you know, I'm going to it, you know, every few minutes just for like idea generation to check something that I'm thinking to get better wording on something that I'm thinking.

20:07And that is really like the worst case scenario for a Google and we are journalists and we cover tech. So it's not surprising that we're sort of early adopters to using this. I don't know how that rolls out to the mainstream, but that was sort of one of the biggest fears going into the Google quarter is that search queries would be hurt or you would see any kind of softness as people use chatbots more. And that hasn't happened at all. I mean, they also, they beat like crazy earnings per share$1.89 versus 151 expected. And then they beat on revenue as well. And I think every, every category beat.

20:45Um, so yeah, I'm curious, I'm curious why you think Google's, I mean, the market reacted so positively to Google. I think it's up 10 % today. Between Google and Microsoft, I was getting texts from people being like, thank God they saved the market. Yeah, it's true. It goes back to that point I made at the very top. If these companies performed really badly, it would take the whole market with them. And Apple will be another important test. NVIDIA, of course, coming up. But when we look at Alphabet, excuse me, it was that dividend, I think, that was so important that investors have been looking for.

21:19And in the lead up to earnings, I was saying, well, there's a small chance, but they absolutely can do a dividend. I don't know what they're waiting for. I asked Ruth Poore at that every single quarter when I got on the phone with her, like, what's your capital allocation plan? How are you thinking about it? Will that change? And, you know, them announcing the dividend is just such an important thing for investors. It was 20 cents a share. It's very small, but this idea that they've reached a new phase, that they're shareholder friendly, that they're returning capital to investors as a sign of maturity at the big tech companies that I think investors have been looking for for a long time and just, I think, indicates that they're thinking about the shareholder more and they're balancing growth and value a little bit more, which shouldn't be all that surprising.

22:04But them doing it is like a big indication that maybe they've entered a new phase. In terms of generative AI not killing or materially hurting search anytime soon, Maybe I'll highlight a few comments from the earnings call last night. Senator Pichai, the CEO, said that they're seeing an increase in search usage among people who use their new AI overviews. And then Philip Schindler, a top exec there, said that advertising is actually finding a place in their Gen AI products. And he said that he was really confident that the opportunity will expand. So it's not like either or. They're finding opportunity.

22:41And I think the key people who aren't like us that have embraced these chatbots early on that are still going to Google search, they're discovering generative AI queries through Google. And that is like so much more important than us, you know, debating Claude or Gemini or GPT. Right. It's just the distribution, I think, is what matters. Right. And that's like what you saw with Meta is they're able to ship their AI model. And all of a sudden, you know, it's in billions of people's hands where OpenAI was celebrating getting in a hundred. I mean, recently so getting in 100 million people's hands in a month now meta has that scale advantage and it's like it's mark versus sam and mark tends to win these battles especially if you look at you know the other victims in the in the way you know evan spiegel and tiktok and all that so it's the distribution that guy's a killer i feel like it's gonna be a good battle i love it i think you know well you know we never got to see zuck versus elon i doubt we're gonna see mma between mean, Zuck and Sam, but this business battle itself is going to be fascinating to watch because it is like, you know, people like we're framing a Sam versus Sundar or Sam versus, I don't know, Dario, but him versus Zuck, I think is going to be very, very interesting to watch.

23:56That's a cage match. I would pay 20 bucks a month for. Definitely. I would too. One last word on Pichai. It's so interesting that like he doesn't enter that conversation, right? It's Musk versus Sam Altman, Musk versus Elon Musk or maybe Dario. And Senator Pichai, he's just sort of never been seen as this wartime CEO, right? And I always question that because I think he's shown flashes of it last week when you had the sit-in, right? And they immediately fired the 28 employees and he came out with a memo the next day that said, this is a business in no uncertain terms. It's interesting. He takes a much more quiet route.

24:36And even you ask me why investors were so stoked on the quarter last night, they're spending a ton in CapEx also. Like the same, if not more as Meta. But then he also talks about engineering durable growth. I hate that phrase so much. It just means that they're going to continue to be efficient. Like Zuckerberg knows how to talk to Wall Street. Maybe he goes, a year of efficiency is a lot more catchy than engineer durable growth. But Senator Pachai and his team are keeping an eye on efficiency while investing huge amounts of money into generative AI while issuing a dividend. So he kind of like hit everything last night.

25:14It's like an investor's dream. Totally. It's a great point of pull out. It does seem like Sundar is like taking a very intentional path right now to sort of revamp like what he's seen as. And I think that like maybe he fired, I think it ended up being 50 people to sort of, because you can't really take over your cloud CEO's office in the middle of a workday and expect to keep your job, especially when you're not being paid to take over your cloud CEO's office. But it also sort of, I think, pretty intentional that, you know, in the past, he let a lot of this go. There were protests over military technology in the past at Google.

25:48And he sort of, you know, said, all right, like, we're going to give you what you want in some ways and talked about AI principles and stuff. And now he's like, all right, we're done with that. And I think he needs to show that, especially given how like the narrative on Google shifted, you know, as they were seen as like a company, it was slow. People were asking if Sundar was the right person. I kind of think that that eventually reached him. And, and some people advising him were like, all right, you got to change the way that this company is operating more urgency, less bullshit and away, away you go.

26:17And to combine that, and he even wrote about it with this like new mission statement, which Alex Heath and I spoke about last week but you combine that and you combine you combine that with better earnings growth and cloud and maybe the fact that you know gem like you're right these gemini and even though i have you know i can sing college praises all day long it's not replacing google search and then you get a good quarter out of google and all of a sudden everything shifts real quickly for them here's an idea maybe the year of zach is ending and the year of sundar pichai it could happen Wouldn't that be interesting?

26:53We've seen, I mean, we've seen comeback stories like that in tech. And so who knows, like, you know, maybe we're going to see that with Google. One person pointed out to me that if Sundar serves like maybe another year or two, he's going to be the longest tenured Google CEO ever. And we'll see if he actually, I mean, there have been some questions about him lasting much longer. It seems like he's pretty solidified, though, in that position. And these recent moves have definitely helped him. I think so too. And plus the people calling for him have been like kind of the loud people on X or Twitter.