November 6, 2025 ā¢ 29 mins
Believe it or not, Google and Apple are undervalued. We discuss Google's aggressive AI investments and the potential to triple its market cap to $10 trillion, alongside Apple's cautious yet strategic approach leveraging local inference capabilities.
We explore the historical contexts, evolving relationships, and the investment potential that could defy market expectations. Be sure to subscribe and share!
------
š LIMITLESS HQ: LISTEN & FOLLOW HERE ā¬ļø
https://limitless.bankless.com/
https://x.com/LimitlessFT
Substack:
https://limitlessft.substack.com/
------
TIMESTAMPS
0:00 Google and Apple: A Tale of Two Giants
0:51 The Bull Case for Google
15:45 The Disconnect with Apple
24:16 Appleās Strategic AI Partnerships
27:42 Conclusion: Value Amidst Market Sentiment
------
RESOURCES
Josh: https://x.com/JoshKale
Ejaaz: https://x.com/cryptopunk7213
------
Not financial or tax advice. See our investment disclosures here:
https://www.bankless.com/disclosuresā
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Ejaaz:
There are two companies currently worth trillions of dollars that I think are (00:03):
undefined
Ejaaz: the most undervalued companies in the world right now. (00:07):
undefined
Ejaaz: Google and Apple are a tale of two tech giants that have taken a very different approach to AI. (00:10):
undefined
Ejaaz: Google, on one hand, has poured billions of dollars to create some of the world's (00:16):
undefined
Ejaaz: leading AI models, going from a corporate has-been to one of the titans of AI. (00:20):
undefined
Ejaaz: They have their own custom GPUs rivaling NVIDIA. They have one of the largest (00:25):
undefined
Ejaaz: user bases in the world with Gmail, Android, YouTube. I could go on. (00:30):
undefined
Ejaaz: Apple, on the other hand, has taken the opposite approach. They haven't invested (00:35):
undefined
Ejaaz: in AI at all. They've been late to update their AI assistant Siri. (00:38):
undefined
Ejaaz: They haven't spent any money to hire AI geniuses like Zuckerberg at Meta. (00:42):
undefined
Ejaaz: But despite this, I think these two companies are poised to be the biggest companies in the world. (00:47):
undefined
Ejaaz: So in this episode, we're going to give you the bull case for Google and Apple (00:52):
undefined
Ejaaz: revealing not just why you should buy their stocks, but also why these two companies (00:56):
undefined
Ejaaz: in particular are very unlikely allies. (01:00):
undefined
Josh: Yeah, I think this episode is going to surprise a lot of people because there's (01:03):
undefined
Josh: this gross disconnect between public market sentiment and public sentiment where (01:07):
undefined
Josh: stocks are trading very high and people don't quite agree with that. (01:11):
undefined
Josh: So I think in this episode, we have a lot of what most would have perceived (01:14):
undefined
Josh: to be contrarian takes about where these companies have come from and where (01:18):
undefined
Josh: they are headed to and how we are actually much more optimistic than I think (01:21):
undefined
Josh: a lot of people think on these companies. (01:24):
undefined
Josh: So Ejaz, maybe you could help us start with the bull case for Google, (01:26):
undefined
Josh: which is one of the first companies we're going to be talking about here. (01:29):
undefined
Ejaaz: What if I told you that Google currently worth $3.4 trillion is an absolute bargain buy right now? (01:32):
undefined
Josh: That's a crazy take. (01:38):
undefined
Ejaaz: You call me crazy. But here's why I think Google is actually worth $10 trillion (01:39):
undefined
Ejaaz: as an AI giant currently masquerading as a simple, humble search engine. (01:44):
undefined
Ejaaz: I've got four reasons for you. (01:50):
undefined
Ejaaz: Number one, Google has the best AI models. (01:51):
undefined
Ejaaz: You have Google Gemini, which is currently crushing across all benchmarks and (01:54):
undefined
Ejaaz: beating ChatGPT on so many different fronts. (01:58):
undefined
Ejaaz: You have VO3, which is their text-to-video model, which is creating cinematic (02:00):
undefined
Ejaaz: masterpieces in a matter of seconds. (02:04):
undefined
Ejaaz: You have Nano Banana, their image generation model, which not just creates really (02:06):
undefined
Ejaaz: good images, but also allows you to edit them. (02:11):
undefined
Ejaaz: And finally, who can forget about Google's Alpha Gemini series, (02:13):
undefined
Ejaaz: their science AI models, which have won not one, but two Nobel Prizes. (02:17):
undefined
Ejaaz: Number two, they're the only real competitor that NVIDIA has. (02:21):
undefined
Ejaaz: They've created TPUs, which is their version of GPUs built to make their own (02:25):
undefined
Ejaaz: custom AI models specifically for Google's software stack and hardware stack. (02:30):
undefined
Ejaaz: They also have some of the craziest distribution of users. They have the number (02:35):
undefined
Ejaaz: one mailing app in Gmail. (02:40):
undefined
Ejaaz: They have the number one entertainment app with YouTube. They dominate search (02:42):
undefined
Ejaaz: engines and advertising. (02:46):
undefined
Ejaaz: And they also have the number one mobile operating system with Android. (02:48):
undefined
Ejaaz: And number four, a little lesser known fact, is they're some of the biggest (02:51):
undefined
Ejaaz: and best investors in the world. (02:55):
undefined
Ejaaz: They currently own 14% of competitor Anthropic and 6% of SpaceX. (02:57):
undefined
Ejaaz: So no matter how way you want to look at it, it's hard to argue why Google isn't (03:03):
undefined
Ejaaz: going to become the most valuable company in the world. (03:08):
undefined
Ejaaz: They have all these strengths, they have the best models, they have the best (03:10):
undefined
Ejaaz: data to train these models, and they have all the compute to build this in-house. (03:13):
undefined
Ejaaz: So I'm struggling to find an argument to see why Google isn't going to be the (03:17):
undefined
Ejaaz: most valuable company within a decade. (03:20):
undefined
Josh: I like looking at history. Google is very much a large player in this Game of (03:22):
undefined
Josh: Thrones that we talk about in the search for AGI. (03:26):
undefined
Josh: And it's helpful to have a little bit of context and pretext to how we got here (03:28):
undefined
Josh: today, where you're saying $10 trillion is a reasonable market cap. (03:31):
undefined
Josh: It's a little outrageous. But there's this funny story associated with the history of Google. (03:34):
undefined
Josh: And it kind of starts in 2019, I believe the year is, where they released a (03:39):
undefined
Josh: paper that was basically the inception of the Transformer. (03:44):
undefined
Josh: And for those who aren't familiar, the Transformer is the piece of technology, (03:47):
undefined
Josh: this little piece of code that has generated and created every single large (03:50):
undefined
Josh: language model that exists today. (03:53):
undefined
Josh: So by all means, Google kind of invented the technology that is running everybody's AI today. (03:55):
undefined
Josh: Naturally, as a result, you would expect them to be the front runner. (04:00):
undefined
Josh: But what they did is they kind of sat on this technology for a very long time. (04:02):
undefined
Josh: So in 2022, when ChatGPT came out, Google actually didn't really have a publicly facing AI product. (04:06):
undefined
Josh: They just had the technology because they were very much a research lab. (04:12):
undefined
Josh: So while OpenAI acquired a million users in the first week, Google is just kind (04:15):
undefined
Josh: of sitting there on, at this time, three to four year old technology that they (04:19):
undefined
Josh: just haven't actually implemented in the product. (04:23):
undefined
Josh: So what they did is they rushed to implement this product. and they created a lot of (04:25):
undefined
Josh: AI systems that started with Bard and then recently is what we know today as (04:30):
undefined
Josh: Gemini that just weren't really quite good. (04:34):
undefined
Josh: And you get the idea that the reason they weren't quite good is because the culture was pretty bad. (04:37):
undefined
Josh: In Google, there was a lot of problems with the workforce and of course a lot (04:42):
undefined
Josh: of disconnect where if you asked Bard at the time to generate the founding fathers, (04:47):
undefined
Josh: it would generate a series of minority women, which was just objectively incorrect and wrong. (04:51):
undefined
Josh: And a lot of people took that as, okay, Google actually doesn't have a chance at succeeding in AI. (04:55):
undefined
Josh: And the reality is that wasn't true because that changed when Sergey Brin comes into the picture. (05:01):
undefined
Josh: So we have a post here, Sergey Brin spotted IRL again, and apparently he's working (05:06):
undefined
Josh: on Gemini at Google. And this is true. (05:10):
undefined
Josh: Sergey Brin, for those who don't know, co-founder of Google. (05:12):
undefined
Josh: He was the guy who started from the beginning and he recognized a wave that (05:14):
undefined
Josh: was too important to miss. (05:18):
undefined
Josh: He came back to Google and he started implementing what they did at early stage Google today. (05:19):
undefined
Josh: And now we have Gemini, which is the most powerful AI model in the world, (05:23):
undefined
Josh: they're at least very much at the frontier with everybody else and they're implementing (05:28):
undefined
Josh: a lot of new services they incepted this amazing technology lost the lead and (05:30):
undefined
Josh: now are each as you're telling me they're probably going to get back into the (05:35):
undefined
Josh: lead because that would be the implication with 10 trillion dollar market cap (05:38):
undefined
Ejaaz: Yep and i think there are four key reasons why uh they hold such a a powerful (05:40):
undefined
Ejaaz: moat and why they're super undervalued i want to start off with, (05:46):
undefined
Ejaaz: Talking about their model, and it's not just like an LLM, it is everything else, right? (05:49):
undefined
Ejaaz: They've tackled and are leading every other medium in terms of AI-generated (05:55):
undefined
Ejaaz: content on the video side with VO, on the image side with Nano Banana, (05:59):
undefined
Ejaaz: and on the scientific discovery side with the Alpha Gemini series. (06:03):
undefined
Ejaaz: Just taking a look at this benchmark setup that I have here, (06:07):
undefined
Ejaaz: this is something that we've started to see with every Gemini model release, (06:10):
undefined
Ejaaz: which is they're frequently and consistently at the top, Josh. (06:13):
undefined
Ejaaz: They're the number one LLM, they're the number one coding assistant, (06:17):
undefined
Ejaaz: and they're the number one agent provider as well. (06:20):
undefined
Ejaaz: And I think that this isn't by coincidence. They've worked very hard to kind (06:23):
undefined
Ejaaz: of like restructure Google from this corporate kind of bloat, (06:27):
undefined
Ejaaz: which they typically had up until AI became a running thing, (06:32):
undefined
Ejaaz: into this like frontier AI tech lab that is constantly innovating, (06:36):
undefined
Ejaaz: changing, and adapting their strategies, given whatever the environment they're faced with. (06:39):
undefined
Ejaaz: So they consistently have the best AI models, but it's also worth explaining (06:43):
undefined
Ejaaz: how they've translated this into user dominance. (06:48):
undefined
Ejaaz: So typically how OpenAI and ChatGPT has done this is they've created a model (06:52):
undefined
Ejaaz: and then they've pioneered basically the chat interface, right? (06:56):
undefined
Ejaaz: They then did the same for text to video into Sora. So they've done a really (06:59):
undefined
Ejaaz: good job of surfacing this to new users. (07:04):
undefined
Ejaaz: Now, Google already has these users and they've done a great job of doing the same thing. (07:07):
undefined
Ejaaz: They've integrated AI into their search. So whenever you search anything on (07:11):
undefined
Ejaaz: Google now, you have something called an AI overview, Josh. (07:14):
undefined
Ejaaz: And something that I learned just this morning, which they released from their (07:18):
undefined
Ejaaz: quarterly report last week, is the monetization rate of this AI search, (07:22):
undefined
Ejaaz: which comes up at the top of their. (07:27):
undefined
Ejaaz: Infamous Google search page is monetizing at the same rate as their search links (07:30):
undefined
Ejaaz: themselves, which is just a crazy undertaking, right? (07:36):
undefined
Ejaaz: Because you would think that Google is killing their own monopoly in search (07:39):
undefined
Ejaaz: by doing something like this, but it turns out that it's actually aiding and (07:43):
undefined
Ejaaz: probably going to offer them more profits going forwards. (07:46):
undefined
Ejaaz: The other thing is they have this whole AI productivity suite, (07:50):
undefined
Ejaaz: which they surface all these different models and features. (07:53):
undefined
Ejaaz: So I guess the point that I'm trying to make is Google isn't just a model creator, (07:55):
undefined
Ejaaz: are a product creator they understand their audience both (08:00):
undefined
Ejaaz: from the enterprise side and the consumer side and i (08:03):
undefined
Ejaaz: think it's highly undervalued and very misunderstood (08:05):
undefined
Ejaaz: that at how good google is at doing that job but what's probably surprising (08:08):
undefined
Ejaaz: to most is one of their products comes in the form of hardware uh their equivalent (08:14):
undefined
Ejaaz: of the gpu called the tpu which has added unlikely the most value to Google's market cap? (08:19):
undefined
Josh: Yeah, TPUs are amazing. And I'm going to do my best to try to explain TPU versus (08:27):
undefined
Josh: GPUs because on the surface, it seems a little confusing, but I think we can simplify it quite a bit. (08:30):
undefined
Josh: Ijus, I want you to imagine a GPU being a chef, a chef who is a master chef, (08:35):
undefined
Josh: who can cook anything and can make anything that you imagine. (08:40):
undefined
Josh: But the reality is that you and myself and the rest of the world, (08:42):
undefined
Josh: we only really want sandwiches. (08:46):
undefined
Josh: So while a chef can make a sandwich, it will take a little bit longer to make a sandwich. (08:47):
undefined
Josh: That is what a GPU does. It can cook you anything you like, But the reality (08:51):
undefined
Josh: is most of the world wants sandwiches. (08:54):
undefined
Josh: A TPU, you can imagine, instead of being a general purpose chef, (08:56):
undefined
Josh: is a production line for sandwiches. (08:59):
undefined
Josh: It is a singular line that has all of the ingredients on top of it to hyper-efficiently (09:01):
undefined
Josh: make sandwiches for the whole world. (09:05):
undefined
Josh: And because the world doesn't want salmon, they don't want steak, (09:07):
undefined
Josh: they just want sandwiches, this production line is much more efficient at making sandwiches. (09:10):
undefined
Josh: So while the GPU can do everything, and a lot of people want a GPU, (09:14):
undefined
Josh: people who are training hardcore AI, they just want a TPU. (09:17):
undefined
Josh: It's called a tensor processing unit, and it is this hyper-focused piece of (09:21):
undefined
Josh: hardware built to perform this thing called matrix math, which a lot of AI hardware is built upon. (09:25):
undefined
Josh: So what is so effective about these TPUs? Well, there's two ways of measuring (09:30):
undefined
Josh: GPU, CPU, compute in general. (09:34):
undefined
Josh: It's performance per watt, and it is cost per token. (09:36):
undefined
Josh: So performance per watt is two times that of the NVIDIA A100s, (09:39):
undefined
Josh: which is one of their leading flagships of chips. (09:44):
undefined
Josh: That means that for every watt of energy you apply, you get double the amount of performance back. (09:46):
undefined
Josh: The other one is cost per token. So for every dollar you spend, (09:51):
undefined
Josh: this is how many tokens you can yield. (09:56):
undefined
Josh: And it's about one and a half to two times better performance per token than (09:58):
undefined
Josh: NVIDIA GPUs. So these TPUs are remarkable. (10:01):
undefined
Josh: They work really, really well. And the question I asked myself when I was going (10:04):
undefined
Josh: through this is, well, why does the whole world not use TPUs? (10:08):
undefined
Josh: Well, the first answer is they actually don't sell them. Google just owns the (10:11):
undefined
Josh: TPU. It is a Google invention. (10:15):
undefined
Josh: They own them, they use them for their own stack, and they vertically integrate (10:17):
undefined
Josh: it through the software. so it's hyper-performance. (10:20):
undefined
Josh: And it's kind of like what we see with Apple chips with the M-Series. (10:22):
undefined
Josh: I always reference the M-Series because it's amazing, is that they've built (10:24):
undefined
Josh: this chip hyper-optimized for their exact production workflow. (10:27):
undefined
Josh: And as a result, the cost efficiencies, performance efficiencies, (10:29):
undefined
Josh: every proficiency is through the roof. (10:33):
undefined
Josh: So that's kind of how I would describe the TPU to a layman. That's kind of how (10:35):
undefined
Josh: I learned it myself. And hopefully that description makes sense. (10:38):
undefined
Ejaaz: That makes a lot of sense. Although I have one slight pushback for you, Josh, (10:41):
undefined
Ejaaz: which is as of last week, they have (10:45):
undefined
Ejaaz: started to sell their tpus to a (10:48):
undefined
Ejaaz: little known ai company uh known as anthropic um (10:52):
undefined
Ejaaz: supposedly the deal is marked at 50 billion dollars (10:55):
undefined
Ejaaz: which doubles their compute offering uh (10:58):
undefined
Ejaaz: from revenue earnings on the year uh so if this deal actually goes through this (11:02):
undefined
Ejaaz: will be the first major gpu sale that has sat outside of nvidia's remit so i (11:06):
undefined
Ejaaz: already know jensen's fuming uh at the head but we will see whether Google takes this seriously. (11:13):
undefined
Ejaaz: I mean, to your point, I don't know if Google has the... (11:19):
undefined
Ejaaz: Manufacturing capability or even the ambition to scale TPUs in the way that (11:23):
undefined
Ejaaz: NVIDIA is done with GPUs. (11:27):
undefined
Ejaaz: I see kind of two major constraints. Number one, Google's just focused on so (11:29):
undefined
Ejaaz: many other things and NVIDIA is like hyper-focused on doing this one thing really well. (11:33):
undefined
Ejaaz: And number two, TPUs, as you said, are highly specific and custom. (11:37):
undefined
Ejaaz: So I'm guessing that the TPUs that they're selling to Anthropic in this deal are very specific. (11:41):
undefined
Ejaaz: Maybe that's a niche that they can kind of take forward, but I don't think it (11:47):
undefined
Ejaaz: could grow as large as NVIDIA. (11:50):
undefined
Ejaaz: Do you think I'm wrong or do you think they're knocking trillions of dollars (11:52):
undefined
Ejaaz: of market cap off of NVIDIA soon? (11:55):
undefined
Josh: Yeah, well, there's two cases. There's one in which you believe TPUs are the next form of GPU. (11:57):
undefined
Josh: It's like a natural extension where everyone's going to want to buy these. (12:01):
undefined
Josh: In that case, if they could do these production at scale, well, (12:04):
undefined
Josh: they will, yeah, they can knock off a trillion or two off of NVIDIA's market cap, sure. (12:07):
undefined
Josh: The reality is, is that I don't believe they are direct Apple to Apple's comparison. (12:10):
undefined
Josh: There's a lot of different types of training that NVIDIA GPUs are good for. the per (12:15):
undefined
Josh: per watt of gpus from nvidia are still higher nvidia (12:18):
undefined
Josh: is still innovating faster they have access to the most cutting edge lithography which (12:21):
undefined
Josh: happens only in a few places in taiwan they really (12:25):
undefined
Josh: have this monopoly on the supply chain that google just can't really compete (12:28):
undefined
Josh: with but google can supplement a lot more compute so to the point of jevin's (12:31):
undefined
Josh: paradox where compute will just expand to the emptiness that there is around (12:36):
undefined
Josh: it um there's no supply demand shortage for these processing units, (12:41):
undefined
Josh: whether they be tensor or graphical. (12:45):
undefined
Josh: And if Google can start selling them, that just opens up another pillar for Google. (12:47):
undefined
Josh: I'm not sure it necessarily hurts NVIDIA, but it allows Google to just make (12:51):
undefined
Josh: more money and print more cash. (12:55):
undefined
Josh: And I think that's a big win for a company that really hasn't made money off (12:57):
undefined
Josh: of the infrastructure play yet, but now has a very clear opportunity to do so. (13:01):
undefined
Ejaaz: Yeah, I mean, I'm just looking at this tweet here. Morgan Stanley kind of like (13:05):
undefined
Ejaaz: up their price and prediction for Google stock in 2026. (13:09):
undefined
Ejaaz: But one thing that really caught my eye is the $158 billion backlog that they (13:15):
undefined
Ejaaz: currently have for their cloud offering. (13:21):
undefined
Ejaaz: And it got me thinking that like, Google has always been a powerhouse when it (13:24):
undefined
Ejaaz: has come to like infrastructure provision and they've done so really well with (13:28):
undefined
Ejaaz: GCP, their cloud product. (13:32):
undefined
Ejaaz: You could also argue the same of Microsoft, Azure. (13:34):
undefined
Ejaaz: Both of these companies have signed major deals, Microsoft and Google, (13:37):
undefined
Ejaaz: in the last two weeks for AI-specific stuff, not just training, (13:41):
undefined
Ejaaz: but also inference costs. (13:45):
undefined
Ejaaz: So I see this like growing trend, Josh, of like infrastructure, (13:46):
undefined
Ejaaz: whether it's chips or just kind of like cloud services, getting really in demand (13:50):
undefined
Ejaaz: as we kind of like scale these AI models up. (13:54):
undefined
Ejaaz: So I don't know, I'm super excited to see where this goes. The final thing is (13:57):
undefined
Ejaaz: Google is just everywhere. (14:01):
undefined
Ejaaz: They are the doorstep to the internet and actually the entire house. (14:04):
undefined
Ejaaz: You know, they run Gemini, they're models, they have the TPU chips, (14:08):
undefined
Ejaaz: which we just mentioned, but they also handle 90% of search. (14:12):
undefined
Ejaaz: They run YouTube, the number one entertainment in the world. (14:16):
undefined
Ejaaz: They have Google Maps, they have Android, they power all Android devices. (14:18):
undefined
Ejaaz: It is just a very gluttonous monopoly that Google has that extends way beyond (14:22):
undefined
Ejaaz: just a simple search engine. And the point I wanna make around here is that (14:27):
undefined
Ejaaz: is a crazy positive feedback loop to when you're creating AI models. (14:31):
undefined
Ejaaz: They have all the data, they have all the compute in-house, so they don't have (14:36):
undefined
Ejaaz: to rely on that tiny island in Taiwan and on Nvidia's dominance here. (14:40):
undefined
Ejaaz: And then they have all the apps to surface that through a very monetizable audience. (14:44):
undefined
Ejaaz: We're talking about billions of users here. (14:48):
undefined
Ejaaz: So whereas OpenAI has to kind of build their audience from scratch and they've (14:50):
undefined
Ejaaz: done so very impressively to 800 million weekly active users, (14:53):
undefined
Ejaaz: Google kind of already had that waiting and ready to go. And they've proven (14:57):
undefined
Ejaaz: that, right? In their quarterly earnings, they have 650 million weekly active users. (15:00):
undefined
Ejaaz: I'm guessing the next quarter, that's going to, you know, increase by at least a third. (15:04):
undefined
Ejaaz: So the point is, Google's dominance extends way beyond a search engine. (15:08):
undefined
Ejaaz: I don't think people are very aware of that. And I think that is one of the (15:12):
undefined
Ejaaz: biggest bull cases for their stock going forwards. (15:15):
undefined
Ejaaz: A counter argument that I've heard to this, Josh, is that they just simply won't execute as well. (15:18):
undefined
Ejaaz: I would agree with you, except that the data tells us otherwise. (15:24):
undefined
Ejaaz: If you look at this chart over here, the retention rate, the three-month retention (15:27):
undefined
Ejaaz: rate, specifically for Google Gemini AI models, has been up and to the right. (15:30):
undefined
Ejaaz: It is pretty insane. We're talking about 90% the three-month retention rate (15:35):
undefined
Ejaaz: and six-month retention rate of 85%. That is staggering for any kind of consumer internet product. (15:40):
undefined
Josh: So you just you've laid out a pretty bullish case (15:45):
undefined
Josh: for google and i'm not sure that mismatches the public perception as (15:48):
undefined
Josh: of late i think that's changed a lot people really hated google but (15:51):
undefined
Josh: they're starting to come around see the light it's trading just beneath all-time highs (15:54):
undefined
Josh: and it very vividly reminds me of another company of about the same size a little (15:57):
undefined
Josh: bit bigger that has it's sitting at all-time highs but there is a very like (16:01):
undefined
Josh: gross disconnect between public sentiment and the company and that's apple um (16:06):
undefined
Josh: there's this post on the screen here that says what the hell is apple doing. (16:10):
undefined
Josh: They failed to make a car. They have no AI investments. (16:13):
undefined
Josh: Siri still sucks. Just releasing the same phone over and over again. How can you miss AI? (16:15):
undefined
Josh: How is that even possible? And Sean, to your point, I agree. (16:20):
undefined
Josh: They've actually, they've swung and missed on just about everything they've (16:24):
undefined
Josh: tried recently. And it's been a big disappointment, but the stock is trading at all time highs. (16:27):
undefined
Josh: So where is this disconnect coming from? How can one group of the internet be (16:32):
undefined
Josh: so polarized against the other group? (16:36):
undefined
Josh: And where all the money is going is clearly into Apple because Apple is trading (16:39):
undefined
Josh: like 4 point something trillion dollar market caps. It's a gigantic company. (16:42):
undefined
Josh: There's a few ways to think about this. And one that I really like is around local inference. (16:46):
undefined
Josh: So the way AI works, and granted, this is kind of funny because (16:52):
undefined
Josh: I think Apple accidentally stumbled upon this conclusion, but I'm going to purvey (16:56):
undefined
Josh: it as if this was all by design. (17:00):
undefined
Josh: There's two kind of ways that companies handle compute. There's cloud-based (17:03):
undefined
Josh: compute, and then there's edge inference that happens. (17:06):
undefined
Josh: So one happens in a data center in a remote place. That's what we see all the (17:09):
undefined
Josh: pictures of, these gigantic factories that have a bunch of GPUs. (17:13):
undefined
Josh: And the other happens locally on your device, whether it be your smartphone (17:16):
undefined
Josh: or your laptop or your desktop, whatever device it may be, (17:19):
undefined
Josh: compute happens locally on that device compute on a local device (17:22):
undefined
Josh: is free compute in a cloud costs money so what (17:25):
undefined
Josh: is apple doing well apple just released a new mac (17:29):
undefined
Josh: mini um which we have a post about here the m3 mac ultra studio or i guess that's (17:32):
undefined
Josh: what it's called is the only device which you can run open source state-of-the-art (17:37):
undefined
Josh: models outside of a data center and that seems strange because there's a lot (17:40):
undefined
Josh: of really powerful gpus there's a lot of really powerful computers but the only (17:44):
undefined
Josh: ones that actually allow you to run these are these Macs. (17:48):
undefined
Josh: Why is that? Well, it's because Apple owns the chips, and they've optimized these chips for AI. (17:51):
undefined
Josh: So even though there is an absence of AI currently running on Apple devices, (17:56):
undefined
Josh: they are fully compute capable of running these very impressive models from (18:00):
undefined
Josh: anything on a Mac Studio down to your iPhone, (18:04):
undefined
Josh: which leads us to the bull case of Apple being they are not subject to a lot (18:06):
undefined
Josh: of the swings that we see in public markets, because they're not investing all of the CapEx. (18:12):
undefined
Josh: Now, EJS, if you'll remember, there was a deal struck between Google and Apple in the past. (18:15):
undefined
Josh: This is not the first time they've worked together. And that was for Google search. (18:20):
undefined
Josh: Back in the day, I think we have a post somewhere about this. (18:23):
undefined
Josh: Yeah, here's a post from EJS. You dug this up from 2018, I think. (18:26):
undefined
Josh: How much is your data worth? So much that Google just paid Apple $9 billion (18:30):
undefined
Josh: to be the default search engine in the iPhone. $9 billion to get people... (18:33):
undefined
Josh: To use a free search. Well, that number has actually jumped to 20 billion as (18:38):
undefined
Josh: of as late as 2022 was the most confirmed number. (18:42):
undefined
Josh: And it's just an outrageous amount of money to spend. But I don't think this (18:45):
undefined
Josh: is the last time they're going to work with each other. (18:48):
undefined
Josh: So Ejas, this week we got some news that there is a new deal on the table. (18:49):
undefined
Josh: And this isn't a search related deal. (18:53):
undefined
Josh: This is instead an AI related deal. (18:55):
undefined
Ejaaz: Yeah, this is like Google and Apple deal 2.0. And it's so much bigger. (18:58):
undefined
Ejaaz: And it's back better than ever, right? So this week, we had news break that (19:03):
undefined
Ejaaz: Apple has asked Google to create their own Gemini model just for Apple users (19:09):
undefined
Ejaaz: that will run on their private cloud. (19:14):
undefined
Ejaaz: Josh, can you break this down for me? What's going on here? (19:17):
undefined
Josh: This is amazingly important. So everybody and their mother is spending trillions (19:19):
undefined
Josh: of dollars of capital building AI data centers. (19:24):
undefined
Josh: OpenAI is spending trillions, Google, Microsoft, name anybody they're spending (19:27):
undefined
Josh: trillions of dollars. There's a question that we frequently ask ourselves on (19:30):
undefined
Josh: the show and public markets are asking, is there a bubble? When is this bubble going to burst? (19:33):
undefined
Josh: How much CapEx is appropriate before you just can't justify the returns on your investment? (19:38):
undefined
Josh: Apple, in a way, in a funny messed up roundabout way, is completely immune to (19:43):
undefined
Josh: these swings, to this capital expenditure, and to this bubble exposure because (19:48):
undefined
Josh: they have not invested much money into building out the resources needed to (19:52):
undefined
Josh: train their own supermodels. (19:55):
undefined
Josh: What they're doing now is a very clear and obvious answer. is they're actually (19:57):
undefined
Josh: offloading that responsibility, offloading the financial risk to Google. (20:01):
undefined
Josh: And they're going to pay them X amount of dollars in order for Google Gemini (20:04):
undefined
Josh: to create a custom model for the Apple ecosystem. (20:08):
undefined
Josh: And I think this is a really powerful thing because we mentioned a little bit (20:10):
undefined
Josh: earlier, there's cloud and then there's local compute. (20:14):
undefined
Josh: And the local compute is actually sufficient enough to create a custom model. (20:16):
undefined
Josh: Don't suffice for a lot of requests that people have. I think a lot of prompts, (20:21):
undefined
Josh: and we talked about this a lot where with cutting edge models, (20:24):
undefined
Josh: I'm not even sure how to really test them because just my requests, (20:27):
undefined
Josh: my needs from an AI model aren't cutting edge. (20:30):
undefined
Josh: They don't need massive amounts of compute or tokens or resources to solve my problems. (20:32):
undefined
Josh: A lot of the times I'm just like, I need help navigating my day. (20:36):
undefined
Josh: I have a lot going on. I need to know like what, what times am I able to do things? (20:39):
undefined
Josh: And the thing with Apple is it actually has all this intelligence. (20:43):
undefined
Josh: And this was the promise with apple intelligence we just never got um but by (20:46):
undefined
Josh: all floating into gemini we start to see these unlocks where apple's the only (20:50):
undefined
Josh: one that has more context than open ai and it does we think about memory a lot (20:54):
undefined
Josh: and context and how powerful it is for a system and what device what system (20:58):
undefined
Josh: have you been using longer than chat gpt that has more info about you what's your smartphone (21:02):
undefined
Josh: and a lot of people will be reluctant to give up that data but in the case that (21:06):
undefined
Josh: it is private and it runs locally and it's able to take all the context from (21:10):
undefined
Josh: your device that creates this really powerful edge compute. (21:14):
undefined
Josh: And that's kind of part one of the bull case for Apple. So I'll stop there to (21:16):
undefined
Josh: see if you disagree, agree. This is like, we're getting into the weeds here a little bit. (21:20):
undefined
Ejaaz: I mean, as you know, I've been one of the biggest bears against Apple for almost a year now. (21:24):
undefined
Josh: And rightfully so. (21:29):
undefined
Ejaaz: Yeah, yeah. I think they've done a terrible job. But as you pointed out, this works both ways. (21:31):
undefined
Ejaaz: They've effectively shielded themselves from the crazy fluctuations that you (21:36):
undefined
Ejaaz: can see in AI, specifically in CapEx, right? So they're not investing trillions of dollars. (21:41):
undefined
Ejaaz: But also when that rollercoaster ride starts going down, they're shielded, (21:45):
undefined
Ejaaz: they're affected from it and they can pick and choose the winners. (21:49):
undefined
Ejaaz: There's also the argument that like all this AI CapEx investment is kind of (21:51):
undefined
Ejaaz: a race to the bottom because whoever gets to AGI first is gonna absolutely disrupt (21:54):
undefined
Ejaaz: and destroy any other competitor. (21:59):
undefined
Ejaaz: An argument that you could potentially give. The other side of it is, (22:01):
undefined
Ejaaz: Josh, I really like your point around the personalization and localization and (22:05):
undefined
Ejaaz: the importance of privacy of having your model on your own device. (22:09):
undefined
Ejaaz: Who cares if there's a general query that can be answered by some research professor versus an AI model? (22:13):
undefined
Ejaaz: I want it to be able to read my text and understand the next thing I'm going to text. (22:19):
undefined
Ejaaz: I want it to be able to download the apps ahead of me even knowing about the (22:23):
undefined
Ejaaz: apps and do the things for me. That's only going to work on personalized devices. (22:26):
undefined
Ejaaz: And as you said, Apple has the distribution. (22:30):
undefined
Josh: Yeah, the use cases for local inference are kind of remarkable when you think (22:33):
undefined
Josh: about having access to all of your texts or emails. (22:36):
undefined
Josh: There's a second thing to this also, which is cost. (22:39):
undefined
Josh: If you remember in a previous episode, we covered Grok Nano, (22:42):
undefined
Josh: which was this very lightweight, very easy to use, very low cost model, (22:45):
undefined
Josh: and how within days of it being released, it went to the very top of all the (22:50):
undefined
Josh: charts in terms of usage. (22:54):
undefined
Josh: Every developer in the world wanted to use it because it was fast and it was cheap. (22:55):
undefined
Josh: And one of the trends that we see as AI develops over time is this race to the bottom. (22:59):
undefined
Josh: The cost per token, the race to zero is fairly aggressive. (23:04):
undefined
Josh: And what Apple's done is they've kind of cheated and they've gone from the starting (23:07):
undefined
Josh: line directly to the finish line without doing the work in between because inference (23:11):
undefined
Josh: on a local device that has an M-series chip or an A-series chip that Apple has is completely free. (23:15):
undefined
Josh: It costs $0 to run a query on these local devices. (23:20):
undefined
Josh: And there's billions of them that are capable of doing this. (23:24):
undefined
Josh: So if you're a developer who wants to build a great experience for a customer, (23:26):
undefined
Josh: there is absolutely no reason why you shouldn't be building for Apple devices (23:30):
undefined
Josh: because the inference costs are free. (23:33):
undefined
Josh: And a lot of these developers who are building applications for the world of (23:36):
undefined
Josh: AI, they're bringing these servers. They're constantly looking for the next (23:38):
undefined
Josh: cheapest one because they want to keep the cost down. (23:41):
undefined
Josh: But if you just throw an app in the App Store that runs on these local models (23:44):
undefined
Josh: that Apple will hopefully produce in partnership with Google, (23:47):
undefined
Josh: You suddenly tap into this gigantic context window, this gigantic bay of memory, (23:50):
undefined
Josh: and it costs you absolutely zero dollars to do. (23:54):
undefined
Josh: And I think market forces will kind of come into play at some point here where (23:57):
undefined
Josh: just the natural tendency to trend towards lower prices will allow a lot more (24:01):
undefined
Josh: developer adoption and to create the experiences on top. (24:05):
undefined
Josh: So while Apple completely and utterly failed at AI version one with Siri, (24:08):
undefined
Josh: it is an abomination. It is the worst AI product of any major company by far. (24:12):
undefined
Josh: And I don't want to get that twisted. they may have accidentally (24:16):
undefined
Josh: just by the fact of vertically integrating these chips stumbled into the best (24:20):
undefined
Josh: case scenario where everybody's spending a bunch of money everyone is trying (24:25):
undefined
Josh: to build the ai first get the user second apple has all the users and they have (24:29):
undefined
Josh: all the data they just need the ai to plug into it and google could be that person to do so yeah (24:34):
undefined
Ejaaz: It's super smart i think the market is finally recognizing that if i were to (24:39):
undefined
Ejaaz: boil all of this down to its core i would say that Apple is actually doing what (24:42):
undefined
Ejaaz: they've always done the best, (24:47):
undefined
Ejaaz: which is create the best consumer experience and products. (24:50):
undefined
Ejaaz: And so rather than jump into the infrastructure wars and the infrastructure (24:54):
undefined
Ejaaz: race in AI with GPUs, they let everyone else do it. (24:57):
undefined
Ejaaz: They cherry pick the best, and then they integrate it in the best, (25:01):
undefined
Ejaaz: most artful way for their own products and services. (25:05):
undefined
Ejaaz: And that's why they have the addicted audience that they have, right? (25:08):
undefined
Ejaaz: I buy iPhones practically every single well, you're having MacBooks at this point. (25:13):
undefined
Ejaaz: I'm addicted to the Apple brand. And there's a reason behind that because it (25:17):
undefined
Ejaaz: has the best all-in consumer experience. (25:20):
undefined
Ejaaz: And that's what Apple knows that they have. And they can afford to basically (25:23):
undefined
Ejaaz: sign these deals and partnerships with the Googles of the world. (25:26):
undefined
Ejaaz: Another thing that I find interesting about this, Josh, is in this case, (25:30):
undefined
Ejaaz: they're paying Google to create their own AI model for them, (25:34):
undefined
Ejaaz: which seems like an L on the balance sheet for Apple. (25:37):
undefined
Ejaaz: But if you play this out, Gemini and Google are definitely going to build their (25:41):
undefined
Ejaaz: own agent, right, which is going to use a bunch of apps and they need to get (25:46):
undefined
Ejaaz: access to billions of consumers. Who has that? (25:51):
undefined
Ejaaz: Apple. I know Google already has it across an array of their different products (25:55):
undefined
Ejaaz: and apps that we explained earlier on. (25:59):
undefined
Ejaaz: But Apple has that addictive user experience that only operates at the hardware (26:01):
undefined
Ejaaz: level through the cellular device. (26:06):
undefined
Ejaaz: And if they maintain that lead, that could be a case for Google actually paying (26:07):
undefined
Ejaaz: Apple a hell of a lot more money to get access to this. (26:12):
undefined
Ejaaz: Is that fair? Is that crazy? Or have I got this wrong? (26:15):
undefined
Josh: Yeah, at the edge, it makes sense that the search thing happens (26:18):
undefined
Josh: again where okay apple owns the user experience (26:21):
undefined
Josh: google owns the software stack google winds up paying (26:24):
undefined
Josh: apple for the exposure um like i said before (26:27):
undefined
Josh: a lot of these companies they they have the intelligence they don't have the (26:30):
undefined
Josh: products they don't have the interface they don't have the trust or the users (26:34):
undefined
Josh: apple has that and by licensing technology from gemini they exclude themselves (26:36):
undefined
Josh: from all the capex fluctuations and the crazy amount of spending without a promised (26:42):
undefined
Josh: return in exchange for creating a a (26:45):
undefined
Josh: AI experience that is frankly the only one that I really want. (26:49):
undefined
Josh: The one that has all of my context. It can read all of my texts, all of my emails. (26:52):
undefined
Josh: It could be the personal assistant that no one else can be yet. (26:56):
undefined
Josh: And that's remarkably powerful. (26:59):
undefined
Josh: And if I want to outsource my thinking to an impressive AI, I could do that. (27:01):
undefined
Josh: And sure, Apple could probably ping some sort of API through Google or whoever (27:05):
undefined
Josh: they're partnered with, like OpenAI. (27:08):
undefined
Josh: But for most of my tasks, most of my AI needs, I just want a really helpful assistant. (27:10):
undefined
Josh: And no one can provide that better than an AI on my iPhone. (27:15):
undefined
Josh: And I think that's kind of the bull case in summary is as AI is able to remove (27:18):
undefined
Josh: themselves from CapEx, as they're able to implement local inference on all these (27:22):
undefined
Josh: devices for zero dollars, there's a lot of network effects that will probably (27:26):
undefined
Josh: come swinging their way. (27:30):
undefined
Josh: And maybe it was accidental, maybe it was by design, but the long tail effects (27:31):
undefined
Josh: of Apple's position right now seem fairly optimistic. (27:35):
undefined
Josh: And I think that's probably why it's priced sitting right at an all-time high (27:37):
undefined
Josh: right now, because we're not the only ones that think that. (27:40):
undefined
Josh: And that just about wraps it up. (27:42):
undefined
Josh: Why we think Google and Apple are both undervalued while at all-time high. (27:44):
undefined
Josh: And while perception might not always match public market sentiment and i think (27:47):
undefined
Josh: it's just it's important to continue to evaluate these companies like we said (27:51):
undefined
Josh: on yesterday's episode um these things matter a lot and how ai is implemented (27:53):
undefined
Josh: really matters a lot and it's changing (27:57):
undefined
Josh: Architecture of the world around us. So it's this exciting conversation we're (27:59):
undefined
Josh: going to continue having as we go. (28:02):
undefined
Josh: I just, as always, thank you so much for watching the episode. (28:04):
undefined
Josh: We really appreciate all the support. (28:07):
undefined
Josh: Things have been going well. In fact, so well that 85% of the people who watch (28:08):
undefined
Josh: this episode on YouTube are not subscribed, which is a problem. (28:12):
undefined
Josh: So if you are watching this and you are not subscribed, please go ahead, (28:16):
undefined
Josh: click that subscribe button and also listen to us wherever you find your podcast. (28:19):
undefined
Josh: One place we're struggling to grow is on the RSS feed on Spotify because it's (28:23):
undefined
Josh: kind of tough to get people to go there so if you listen and you want to support (28:27):
undefined
Josh: the show please go there subscribe on spotify subscribe on your rss feed wherever (28:30):
undefined
Josh: your player may be apple podcast even if (28:34):
undefined
Ejaaz: You don't want to listen if you're taking a shower just put it on play put it (28:36):
undefined
Ejaaz: in the other room it helps us out a lot. (28:40):
undefined
Josh: That's what my dad does it's great so yeah we just appreciate the support of (28:43):
undefined
Josh: whatever you can give all the comments it's been amazing we're coming off our (28:46):
undefined
Josh: best month ever and we continue to keep our foot on the gas thanks to all the (28:49):
undefined
Josh: support so thank you for watching as always we will see you in the next episode and peace (28:52):
undefined
undefined
Ejaaz: the most undervalued companies in the world right now. (00:07):
undefined
Ejaaz: Google and Apple are a tale of two tech giants that have taken a very different approach to AI. (00:10):
undefined
Ejaaz: Google, on one hand, has poured billions of dollars to create some of the world's (00:16):
undefined
Ejaaz: leading AI models, going from a corporate has-been to one of the titans of AI. (00:20):
undefined
Ejaaz: They have their own custom GPUs rivaling NVIDIA. They have one of the largest (00:25):
undefined
Ejaaz: user bases in the world with Gmail, Android, YouTube. I could go on. (00:30):
undefined
Ejaaz: Apple, on the other hand, has taken the opposite approach. They haven't invested (00:35):
undefined
Ejaaz: in AI at all. They've been late to update their AI assistant Siri. (00:38):
undefined
Ejaaz: They haven't spent any money to hire AI geniuses like Zuckerberg at Meta. (00:42):
undefined
Ejaaz: But despite this, I think these two companies are poised to be the biggest companies in the world. (00:47):
undefined
Ejaaz: So in this episode, we're going to give you the bull case for Google and Apple (00:52):
undefined
Ejaaz: revealing not just why you should buy their stocks, but also why these two companies (00:56):
undefined
Ejaaz: in particular are very unlikely allies. (01:00):
undefined
Josh: Yeah, I think this episode is going to surprise a lot of people because there's (01:03):
undefined
Josh: this gross disconnect between public market sentiment and public sentiment where (01:07):
undefined
Josh: stocks are trading very high and people don't quite agree with that. (01:11):
undefined
Josh: So I think in this episode, we have a lot of what most would have perceived (01:14):
undefined
Josh: to be contrarian takes about where these companies have come from and where (01:18):
undefined
Josh: they are headed to and how we are actually much more optimistic than I think (01:21):
undefined
Josh: a lot of people think on these companies. (01:24):
undefined
Josh: So Ejaz, maybe you could help us start with the bull case for Google, (01:26):
undefined
Josh: which is one of the first companies we're going to be talking about here. (01:29):
undefined
Ejaaz: What if I told you that Google currently worth $3.4 trillion is an absolute bargain buy right now? (01:32):
undefined
Josh: That's a crazy take. (01:38):
undefined
Ejaaz: You call me crazy. But here's why I think Google is actually worth $10 trillion (01:39):
undefined
Ejaaz: as an AI giant currently masquerading as a simple, humble search engine. (01:44):
undefined
Ejaaz: I've got four reasons for you. (01:50):
undefined
Ejaaz: Number one, Google has the best AI models. (01:51):
undefined
Ejaaz: You have Google Gemini, which is currently crushing across all benchmarks and (01:54):
undefined
Ejaaz: beating ChatGPT on so many different fronts. (01:58):
undefined
Ejaaz: You have VO3, which is their text-to-video model, which is creating cinematic (02:00):
undefined
Ejaaz: masterpieces in a matter of seconds. (02:04):
undefined
Ejaaz: You have Nano Banana, their image generation model, which not just creates really (02:06):
undefined
Ejaaz: good images, but also allows you to edit them. (02:11):
undefined
Ejaaz: And finally, who can forget about Google's Alpha Gemini series, (02:13):
undefined
Ejaaz: their science AI models, which have won not one, but two Nobel Prizes. (02:17):
undefined
Ejaaz: Number two, they're the only real competitor that NVIDIA has. (02:21):
undefined
Ejaaz: They've created TPUs, which is their version of GPUs built to make their own (02:25):
undefined
Ejaaz: custom AI models specifically for Google's software stack and hardware stack. (02:30):
undefined
Ejaaz: They also have some of the craziest distribution of users. They have the number (02:35):
undefined
Ejaaz: one mailing app in Gmail. (02:40):
undefined
Ejaaz: They have the number one entertainment app with YouTube. They dominate search (02:42):
undefined
Ejaaz: engines and advertising. (02:46):
undefined
Ejaaz: And they also have the number one mobile operating system with Android. (02:48):
undefined
Ejaaz: And number four, a little lesser known fact, is they're some of the biggest (02:51):
undefined
Ejaaz: and best investors in the world. (02:55):
undefined
Ejaaz: They currently own 14% of competitor Anthropic and 6% of SpaceX. (02:57):
undefined
Ejaaz: So no matter how way you want to look at it, it's hard to argue why Google isn't (03:03):
undefined
Ejaaz: going to become the most valuable company in the world. (03:08):
undefined
Ejaaz: They have all these strengths, they have the best models, they have the best (03:10):
undefined
Ejaaz: data to train these models, and they have all the compute to build this in-house. (03:13):
undefined
Ejaaz: So I'm struggling to find an argument to see why Google isn't going to be the (03:17):
undefined
Ejaaz: most valuable company within a decade. (03:20):
undefined
Josh: I like looking at history. Google is very much a large player in this Game of (03:22):
undefined
Josh: Thrones that we talk about in the search for AGI. (03:26):
undefined
Josh: And it's helpful to have a little bit of context and pretext to how we got here (03:28):
undefined
Josh: today, where you're saying $10 trillion is a reasonable market cap. (03:31):
undefined
Josh: It's a little outrageous. But there's this funny story associated with the history of Google. (03:34):
undefined
Josh: And it kind of starts in 2019, I believe the year is, where they released a (03:39):
undefined
Josh: paper that was basically the inception of the Transformer. (03:44):
undefined
Josh: And for those who aren't familiar, the Transformer is the piece of technology, (03:47):
undefined
Josh: this little piece of code that has generated and created every single large (03:50):
undefined
Josh: language model that exists today. (03:53):
undefined
Josh: So by all means, Google kind of invented the technology that is running everybody's AI today. (03:55):
undefined
Josh: Naturally, as a result, you would expect them to be the front runner. (04:00):
undefined
Josh: But what they did is they kind of sat on this technology for a very long time. (04:02):
undefined
Josh: So in 2022, when ChatGPT came out, Google actually didn't really have a publicly facing AI product. (04:06):
undefined
Josh: They just had the technology because they were very much a research lab. (04:12):
undefined
Josh: So while OpenAI acquired a million users in the first week, Google is just kind (04:15):
undefined
Josh: of sitting there on, at this time, three to four year old technology that they (04:19):
undefined
Josh: just haven't actually implemented in the product. (04:23):
undefined
Josh: So what they did is they rushed to implement this product. and they created a lot of (04:25):
undefined
Josh: AI systems that started with Bard and then recently is what we know today as (04:30):
undefined
Josh: Gemini that just weren't really quite good. (04:34):
undefined
Josh: And you get the idea that the reason they weren't quite good is because the culture was pretty bad. (04:37):
undefined
Josh: In Google, there was a lot of problems with the workforce and of course a lot (04:42):
undefined
Josh: of disconnect where if you asked Bard at the time to generate the founding fathers, (04:47):
undefined
Josh: it would generate a series of minority women, which was just objectively incorrect and wrong. (04:51):
undefined
Josh: And a lot of people took that as, okay, Google actually doesn't have a chance at succeeding in AI. (04:55):
undefined
Josh: And the reality is that wasn't true because that changed when Sergey Brin comes into the picture. (05:01):
undefined
Josh: So we have a post here, Sergey Brin spotted IRL again, and apparently he's working (05:06):
undefined
Josh: on Gemini at Google. And this is true. (05:10):
undefined
Josh: Sergey Brin, for those who don't know, co-founder of Google. (05:12):
undefined
Josh: He was the guy who started from the beginning and he recognized a wave that (05:14):
undefined
Josh: was too important to miss. (05:18):
undefined
Josh: He came back to Google and he started implementing what they did at early stage Google today. (05:19):
undefined
Josh: And now we have Gemini, which is the most powerful AI model in the world, (05:23):
undefined
Josh: they're at least very much at the frontier with everybody else and they're implementing (05:28):
undefined
Josh: a lot of new services they incepted this amazing technology lost the lead and (05:30):
undefined
Josh: now are each as you're telling me they're probably going to get back into the (05:35):
undefined
Josh: lead because that would be the implication with 10 trillion dollar market cap (05:38):
undefined
Ejaaz: Yep and i think there are four key reasons why uh they hold such a a powerful (05:40):
undefined
Ejaaz: moat and why they're super undervalued i want to start off with, (05:46):
undefined
Ejaaz: Talking about their model, and it's not just like an LLM, it is everything else, right? (05:49):
undefined
Ejaaz: They've tackled and are leading every other medium in terms of AI-generated (05:55):
undefined
Ejaaz: content on the video side with VO, on the image side with Nano Banana, (05:59):
undefined
Ejaaz: and on the scientific discovery side with the Alpha Gemini series. (06:03):
undefined
Ejaaz: Just taking a look at this benchmark setup that I have here, (06:07):
undefined
Ejaaz: this is something that we've started to see with every Gemini model release, (06:10):
undefined
Ejaaz: which is they're frequently and consistently at the top, Josh. (06:13):
undefined
Ejaaz: They're the number one LLM, they're the number one coding assistant, (06:17):
undefined
Ejaaz: and they're the number one agent provider as well. (06:20):
undefined
Ejaaz: And I think that this isn't by coincidence. They've worked very hard to kind (06:23):
undefined
Ejaaz: of like restructure Google from this corporate kind of bloat, (06:27):
undefined
Ejaaz: which they typically had up until AI became a running thing, (06:32):
undefined
Ejaaz: into this like frontier AI tech lab that is constantly innovating, (06:36):
undefined
Ejaaz: changing, and adapting their strategies, given whatever the environment they're faced with. (06:39):
undefined
Ejaaz: So they consistently have the best AI models, but it's also worth explaining (06:43):
undefined
Ejaaz: how they've translated this into user dominance. (06:48):
undefined
Ejaaz: So typically how OpenAI and ChatGPT has done this is they've created a model (06:52):
undefined
Ejaaz: and then they've pioneered basically the chat interface, right? (06:56):
undefined
Ejaaz: They then did the same for text to video into Sora. So they've done a really (06:59):
undefined
Ejaaz: good job of surfacing this to new users. (07:04):
undefined
Ejaaz: Now, Google already has these users and they've done a great job of doing the same thing. (07:07):
undefined
Ejaaz: They've integrated AI into their search. So whenever you search anything on (07:11):
undefined
Ejaaz: Google now, you have something called an AI overview, Josh. (07:14):
undefined
Ejaaz: And something that I learned just this morning, which they released from their (07:18):
undefined
Ejaaz: quarterly report last week, is the monetization rate of this AI search, (07:22):
undefined
Ejaaz: which comes up at the top of their. (07:27):
undefined
Ejaaz: Infamous Google search page is monetizing at the same rate as their search links (07:30):
undefined
Ejaaz: themselves, which is just a crazy undertaking, right? (07:36):
undefined
Ejaaz: Because you would think that Google is killing their own monopoly in search (07:39):
undefined
Ejaaz: by doing something like this, but it turns out that it's actually aiding and (07:43):
undefined
Ejaaz: probably going to offer them more profits going forwards. (07:46):
undefined
Ejaaz: The other thing is they have this whole AI productivity suite, (07:50):
undefined
Ejaaz: which they surface all these different models and features. (07:53):
undefined
Ejaaz: So I guess the point that I'm trying to make is Google isn't just a model creator, (07:55):
undefined
Ejaaz: are a product creator they understand their audience both (08:00):
undefined
Ejaaz: from the enterprise side and the consumer side and i (08:03):
undefined
Ejaaz: think it's highly undervalued and very misunderstood (08:05):
undefined
Ejaaz: that at how good google is at doing that job but what's probably surprising (08:08):
undefined
Ejaaz: to most is one of their products comes in the form of hardware uh their equivalent (08:14):
undefined
Ejaaz: of the gpu called the tpu which has added unlikely the most value to Google's market cap? (08:19):
undefined
Josh: Yeah, TPUs are amazing. And I'm going to do my best to try to explain TPU versus (08:27):
undefined
Josh: GPUs because on the surface, it seems a little confusing, but I think we can simplify it quite a bit. (08:30):
undefined
Josh: Ijus, I want you to imagine a GPU being a chef, a chef who is a master chef, (08:35):
undefined
Josh: who can cook anything and can make anything that you imagine. (08:40):
undefined
Josh: But the reality is that you and myself and the rest of the world, (08:42):
undefined
Josh: we only really want sandwiches. (08:46):
undefined
Josh: So while a chef can make a sandwich, it will take a little bit longer to make a sandwich. (08:47):
undefined
Josh: That is what a GPU does. It can cook you anything you like, But the reality (08:51):
undefined
Josh: is most of the world wants sandwiches. (08:54):
undefined
Josh: A TPU, you can imagine, instead of being a general purpose chef, (08:56):
undefined
Josh: is a production line for sandwiches. (08:59):
undefined
Josh: It is a singular line that has all of the ingredients on top of it to hyper-efficiently (09:01):
undefined
Josh: make sandwiches for the whole world. (09:05):
undefined
Josh: And because the world doesn't want salmon, they don't want steak, (09:07):
undefined
Josh: they just want sandwiches, this production line is much more efficient at making sandwiches. (09:10):
undefined
Josh: So while the GPU can do everything, and a lot of people want a GPU, (09:14):
undefined
Josh: people who are training hardcore AI, they just want a TPU. (09:17):
undefined
Josh: It's called a tensor processing unit, and it is this hyper-focused piece of (09:21):
undefined
Josh: hardware built to perform this thing called matrix math, which a lot of AI hardware is built upon. (09:25):
undefined
Josh: So what is so effective about these TPUs? Well, there's two ways of measuring (09:30):
undefined
Josh: GPU, CPU, compute in general. (09:34):
undefined
Josh: It's performance per watt, and it is cost per token. (09:36):
undefined
Josh: So performance per watt is two times that of the NVIDIA A100s, (09:39):
undefined
Josh: which is one of their leading flagships of chips. (09:44):
undefined
Josh: That means that for every watt of energy you apply, you get double the amount of performance back. (09:46):
undefined
Josh: The other one is cost per token. So for every dollar you spend, (09:51):
undefined
Josh: this is how many tokens you can yield. (09:56):
undefined
Josh: And it's about one and a half to two times better performance per token than (09:58):
undefined
Josh: NVIDIA GPUs. So these TPUs are remarkable. (10:01):
undefined
Josh: They work really, really well. And the question I asked myself when I was going (10:04):
undefined
Josh: through this is, well, why does the whole world not use TPUs? (10:08):
undefined
Josh: Well, the first answer is they actually don't sell them. Google just owns the (10:11):
undefined
Josh: TPU. It is a Google invention. (10:15):
undefined
Josh: They own them, they use them for their own stack, and they vertically integrate (10:17):
undefined
Josh: it through the software. so it's hyper-performance. (10:20):
undefined
Josh: And it's kind of like what we see with Apple chips with the M-Series. (10:22):
undefined
Josh: I always reference the M-Series because it's amazing, is that they've built (10:24):
undefined
Josh: this chip hyper-optimized for their exact production workflow. (10:27):
undefined
Josh: And as a result, the cost efficiencies, performance efficiencies, (10:29):
undefined
Josh: every proficiency is through the roof. (10:33):
undefined
Josh: So that's kind of how I would describe the TPU to a layman. That's kind of how (10:35):
undefined
Josh: I learned it myself. And hopefully that description makes sense. (10:38):
undefined
Ejaaz: That makes a lot of sense. Although I have one slight pushback for you, Josh, (10:41):
undefined
Ejaaz: which is as of last week, they have (10:45):
undefined
Ejaaz: started to sell their tpus to a (10:48):
undefined
Ejaaz: little known ai company uh known as anthropic um (10:52):
undefined
Ejaaz: supposedly the deal is marked at 50 billion dollars (10:55):
undefined
Ejaaz: which doubles their compute offering uh (10:58):
undefined
Ejaaz: from revenue earnings on the year uh so if this deal actually goes through this (11:02):
undefined
Ejaaz: will be the first major gpu sale that has sat outside of nvidia's remit so i (11:06):
undefined
Ejaaz: already know jensen's fuming uh at the head but we will see whether Google takes this seriously. (11:13):
undefined
Ejaaz: I mean, to your point, I don't know if Google has the... (11:19):
undefined
Ejaaz: Manufacturing capability or even the ambition to scale TPUs in the way that (11:23):
undefined
Ejaaz: NVIDIA is done with GPUs. (11:27):
undefined
Ejaaz: I see kind of two major constraints. Number one, Google's just focused on so (11:29):
undefined
Ejaaz: many other things and NVIDIA is like hyper-focused on doing this one thing really well. (11:33):
undefined
Ejaaz: And number two, TPUs, as you said, are highly specific and custom. (11:37):
undefined
Ejaaz: So I'm guessing that the TPUs that they're selling to Anthropic in this deal are very specific. (11:41):
undefined
Ejaaz: Maybe that's a niche that they can kind of take forward, but I don't think it (11:47):
undefined
Ejaaz: could grow as large as NVIDIA. (11:50):
undefined
Ejaaz: Do you think I'm wrong or do you think they're knocking trillions of dollars (11:52):
undefined
Ejaaz: of market cap off of NVIDIA soon? (11:55):
undefined
Josh: Yeah, well, there's two cases. There's one in which you believe TPUs are the next form of GPU. (11:57):
undefined
Josh: It's like a natural extension where everyone's going to want to buy these. (12:01):
undefined
Josh: In that case, if they could do these production at scale, well, (12:04):
undefined
Josh: they will, yeah, they can knock off a trillion or two off of NVIDIA's market cap, sure. (12:07):
undefined
Josh: The reality is, is that I don't believe they are direct Apple to Apple's comparison. (12:10):
undefined
Josh: There's a lot of different types of training that NVIDIA GPUs are good for. the per (12:15):
undefined
Josh: per watt of gpus from nvidia are still higher nvidia (12:18):
undefined
Josh: is still innovating faster they have access to the most cutting edge lithography which (12:21):
undefined
Josh: happens only in a few places in taiwan they really (12:25):
undefined
Josh: have this monopoly on the supply chain that google just can't really compete (12:28):
undefined
Josh: with but google can supplement a lot more compute so to the point of jevin's (12:31):
undefined
Josh: paradox where compute will just expand to the emptiness that there is around (12:36):
undefined
Josh: it um there's no supply demand shortage for these processing units, (12:41):
undefined
Josh: whether they be tensor or graphical. (12:45):
undefined
Josh: And if Google can start selling them, that just opens up another pillar for Google. (12:47):
undefined
Josh: I'm not sure it necessarily hurts NVIDIA, but it allows Google to just make (12:51):
undefined
Josh: more money and print more cash. (12:55):
undefined
Josh: And I think that's a big win for a company that really hasn't made money off (12:57):
undefined
Josh: of the infrastructure play yet, but now has a very clear opportunity to do so. (13:01):
undefined
Ejaaz: Yeah, I mean, I'm just looking at this tweet here. Morgan Stanley kind of like (13:05):
undefined
Ejaaz: up their price and prediction for Google stock in 2026. (13:09):
undefined
Ejaaz: But one thing that really caught my eye is the $158 billion backlog that they (13:15):
undefined
Ejaaz: currently have for their cloud offering. (13:21):
undefined
Ejaaz: And it got me thinking that like, Google has always been a powerhouse when it (13:24):
undefined
Ejaaz: has come to like infrastructure provision and they've done so really well with (13:28):
undefined
Ejaaz: GCP, their cloud product. (13:32):
undefined
Ejaaz: You could also argue the same of Microsoft, Azure. (13:34):
undefined
Ejaaz: Both of these companies have signed major deals, Microsoft and Google, (13:37):
undefined
Ejaaz: in the last two weeks for AI-specific stuff, not just training, (13:41):
undefined
Ejaaz: but also inference costs. (13:45):
undefined
Ejaaz: So I see this like growing trend, Josh, of like infrastructure, (13:46):
undefined
Ejaaz: whether it's chips or just kind of like cloud services, getting really in demand (13:50):
undefined
Ejaaz: as we kind of like scale these AI models up. (13:54):
undefined
Ejaaz: So I don't know, I'm super excited to see where this goes. The final thing is (13:57):
undefined
Ejaaz: Google is just everywhere. (14:01):
undefined
Ejaaz: They are the doorstep to the internet and actually the entire house. (14:04):
undefined
Ejaaz: You know, they run Gemini, they're models, they have the TPU chips, (14:08):
undefined
Ejaaz: which we just mentioned, but they also handle 90% of search. (14:12):
undefined
Ejaaz: They run YouTube, the number one entertainment in the world. (14:16):
undefined
Ejaaz: They have Google Maps, they have Android, they power all Android devices. (14:18):
undefined
Ejaaz: It is just a very gluttonous monopoly that Google has that extends way beyond (14:22):
undefined
Ejaaz: just a simple search engine. And the point I wanna make around here is that (14:27):
undefined
Ejaaz: is a crazy positive feedback loop to when you're creating AI models. (14:31):
undefined
Ejaaz: They have all the data, they have all the compute in-house, so they don't have (14:36):
undefined
Ejaaz: to rely on that tiny island in Taiwan and on Nvidia's dominance here. (14:40):
undefined
Ejaaz: And then they have all the apps to surface that through a very monetizable audience. (14:44):
undefined
Ejaaz: We're talking about billions of users here. (14:48):
undefined
Ejaaz: So whereas OpenAI has to kind of build their audience from scratch and they've (14:50):
undefined
Ejaaz: done so very impressively to 800 million weekly active users, (14:53):
undefined
Ejaaz: Google kind of already had that waiting and ready to go. And they've proven (14:57):
undefined
Ejaaz: that, right? In their quarterly earnings, they have 650 million weekly active users. (15:00):
undefined
Ejaaz: I'm guessing the next quarter, that's going to, you know, increase by at least a third. (15:04):
undefined
Ejaaz: So the point is, Google's dominance extends way beyond a search engine. (15:08):
undefined
Ejaaz: I don't think people are very aware of that. And I think that is one of the (15:12):
undefined
Ejaaz: biggest bull cases for their stock going forwards. (15:15):
undefined
Ejaaz: A counter argument that I've heard to this, Josh, is that they just simply won't execute as well. (15:18):
undefined
Ejaaz: I would agree with you, except that the data tells us otherwise. (15:24):
undefined
Ejaaz: If you look at this chart over here, the retention rate, the three-month retention (15:27):
undefined
Ejaaz: rate, specifically for Google Gemini AI models, has been up and to the right. (15:30):
undefined
Ejaaz: It is pretty insane. We're talking about 90% the three-month retention rate (15:35):
undefined
Ejaaz: and six-month retention rate of 85%. That is staggering for any kind of consumer internet product. (15:40):
undefined
Josh: So you just you've laid out a pretty bullish case (15:45):
undefined
Josh: for google and i'm not sure that mismatches the public perception as (15:48):
undefined
Josh: of late i think that's changed a lot people really hated google but (15:51):
undefined
Josh: they're starting to come around see the light it's trading just beneath all-time highs (15:54):
undefined
Josh: and it very vividly reminds me of another company of about the same size a little (15:57):
undefined
Josh: bit bigger that has it's sitting at all-time highs but there is a very like (16:01):
undefined
Josh: gross disconnect between public sentiment and the company and that's apple um (16:06):
undefined
Josh: there's this post on the screen here that says what the hell is apple doing. (16:10):
undefined
Josh: They failed to make a car. They have no AI investments. (16:13):
undefined
Josh: Siri still sucks. Just releasing the same phone over and over again. How can you miss AI? (16:15):
undefined
Josh: How is that even possible? And Sean, to your point, I agree. (16:20):
undefined
Josh: They've actually, they've swung and missed on just about everything they've (16:24):
undefined
Josh: tried recently. And it's been a big disappointment, but the stock is trading at all time highs. (16:27):
undefined
Josh: So where is this disconnect coming from? How can one group of the internet be (16:32):
undefined
Josh: so polarized against the other group? (16:36):
undefined
Josh: And where all the money is going is clearly into Apple because Apple is trading (16:39):
undefined
Josh: like 4 point something trillion dollar market caps. It's a gigantic company. (16:42):
undefined
Josh: There's a few ways to think about this. And one that I really like is around local inference. (16:46):
undefined
Josh: So the way AI works, and granted, this is kind of funny because (16:52):
undefined
Josh: I think Apple accidentally stumbled upon this conclusion, but I'm going to purvey (16:56):
undefined
Josh: it as if this was all by design. (17:00):
undefined
Josh: There's two kind of ways that companies handle compute. There's cloud-based (17:03):
undefined
Josh: compute, and then there's edge inference that happens. (17:06):
undefined
Josh: So one happens in a data center in a remote place. That's what we see all the (17:09):
undefined
Josh: pictures of, these gigantic factories that have a bunch of GPUs. (17:13):
undefined
Josh: And the other happens locally on your device, whether it be your smartphone (17:16):
undefined
Josh: or your laptop or your desktop, whatever device it may be, (17:19):
undefined
Josh: compute happens locally on that device compute on a local device (17:22):
undefined
Josh: is free compute in a cloud costs money so what (17:25):
undefined
Josh: is apple doing well apple just released a new mac (17:29):
undefined
Josh: mini um which we have a post about here the m3 mac ultra studio or i guess that's (17:32):
undefined
Josh: what it's called is the only device which you can run open source state-of-the-art (17:37):
undefined
Josh: models outside of a data center and that seems strange because there's a lot (17:40):
undefined
Josh: of really powerful gpus there's a lot of really powerful computers but the only (17:44):
undefined
Josh: ones that actually allow you to run these are these Macs. (17:48):
undefined
Josh: Why is that? Well, it's because Apple owns the chips, and they've optimized these chips for AI. (17:51):
undefined
Josh: So even though there is an absence of AI currently running on Apple devices, (17:56):
undefined
Josh: they are fully compute capable of running these very impressive models from (18:00):
undefined
Josh: anything on a Mac Studio down to your iPhone, (18:04):
undefined
Josh: which leads us to the bull case of Apple being they are not subject to a lot (18:06):
undefined
Josh: of the swings that we see in public markets, because they're not investing all of the CapEx. (18:12):
undefined
Josh: Now, EJS, if you'll remember, there was a deal struck between Google and Apple in the past. (18:15):
undefined
Josh: This is not the first time they've worked together. And that was for Google search. (18:20):
undefined
Josh: Back in the day, I think we have a post somewhere about this. (18:23):
undefined
Josh: Yeah, here's a post from EJS. You dug this up from 2018, I think. (18:26):
undefined
Josh: How much is your data worth? So much that Google just paid Apple $9 billion (18:30):
undefined
Josh: to be the default search engine in the iPhone. $9 billion to get people... (18:33):
undefined
Josh: To use a free search. Well, that number has actually jumped to 20 billion as (18:38):
undefined
Josh: of as late as 2022 was the most confirmed number. (18:42):
undefined
Josh: And it's just an outrageous amount of money to spend. But I don't think this (18:45):
undefined
Josh: is the last time they're going to work with each other. (18:48):
undefined
Josh: So Ejas, this week we got some news that there is a new deal on the table. (18:49):
undefined
Josh: And this isn't a search related deal. (18:53):
undefined
Josh: This is instead an AI related deal. (18:55):
undefined
Ejaaz: Yeah, this is like Google and Apple deal 2.0. And it's so much bigger. (18:58):
undefined
Ejaaz: And it's back better than ever, right? So this week, we had news break that (19:03):
undefined
Ejaaz: Apple has asked Google to create their own Gemini model just for Apple users (19:09):
undefined
Ejaaz: that will run on their private cloud. (19:14):
undefined
Ejaaz: Josh, can you break this down for me? What's going on here? (19:17):
undefined
Josh: This is amazingly important. So everybody and their mother is spending trillions (19:19):
undefined
Josh: of dollars of capital building AI data centers. (19:24):
undefined
Josh: OpenAI is spending trillions, Google, Microsoft, name anybody they're spending (19:27):
undefined
Josh: trillions of dollars. There's a question that we frequently ask ourselves on (19:30):
undefined
Josh: the show and public markets are asking, is there a bubble? When is this bubble going to burst? (19:33):
undefined
Josh: How much CapEx is appropriate before you just can't justify the returns on your investment? (19:38):
undefined
Josh: Apple, in a way, in a funny messed up roundabout way, is completely immune to (19:43):
undefined
Josh: these swings, to this capital expenditure, and to this bubble exposure because (19:48):
undefined
Josh: they have not invested much money into building out the resources needed to (19:52):
undefined
Josh: train their own supermodels. (19:55):
undefined
Josh: What they're doing now is a very clear and obvious answer. is they're actually (19:57):
undefined
Josh: offloading that responsibility, offloading the financial risk to Google. (20:01):
undefined
Josh: And they're going to pay them X amount of dollars in order for Google Gemini (20:04):
undefined
Josh: to create a custom model for the Apple ecosystem. (20:08):
undefined
Josh: And I think this is a really powerful thing because we mentioned a little bit (20:10):
undefined
Josh: earlier, there's cloud and then there's local compute. (20:14):
undefined
Josh: And the local compute is actually sufficient enough to create a custom model. (20:16):
undefined
Josh: Don't suffice for a lot of requests that people have. I think a lot of prompts, (20:21):
undefined
Josh: and we talked about this a lot where with cutting edge models, (20:24):
undefined
Josh: I'm not even sure how to really test them because just my requests, (20:27):
undefined
Josh: my needs from an AI model aren't cutting edge. (20:30):
undefined
Josh: They don't need massive amounts of compute or tokens or resources to solve my problems. (20:32):
undefined
Josh: A lot of the times I'm just like, I need help navigating my day. (20:36):
undefined
Josh: I have a lot going on. I need to know like what, what times am I able to do things? (20:39):
undefined
Josh: And the thing with Apple is it actually has all this intelligence. (20:43):
undefined
Josh: And this was the promise with apple intelligence we just never got um but by (20:46):
undefined
Josh: all floating into gemini we start to see these unlocks where apple's the only (20:50):
undefined
Josh: one that has more context than open ai and it does we think about memory a lot (20:54):
undefined
Josh: and context and how powerful it is for a system and what device what system (20:58):
undefined
Josh: have you been using longer than chat gpt that has more info about you what's your smartphone (21:02):
undefined
Josh: and a lot of people will be reluctant to give up that data but in the case that (21:06):
undefined
Josh: it is private and it runs locally and it's able to take all the context from (21:10):
undefined
Josh: your device that creates this really powerful edge compute. (21:14):
undefined
Josh: And that's kind of part one of the bull case for Apple. So I'll stop there to (21:16):
undefined
Josh: see if you disagree, agree. This is like, we're getting into the weeds here a little bit. (21:20):
undefined
Ejaaz: I mean, as you know, I've been one of the biggest bears against Apple for almost a year now. (21:24):
undefined
Josh: And rightfully so. (21:29):
undefined
Ejaaz: Yeah, yeah. I think they've done a terrible job. But as you pointed out, this works both ways. (21:31):
undefined
Ejaaz: They've effectively shielded themselves from the crazy fluctuations that you (21:36):
undefined
Ejaaz: can see in AI, specifically in CapEx, right? So they're not investing trillions of dollars. (21:41):
undefined
Ejaaz: But also when that rollercoaster ride starts going down, they're shielded, (21:45):
undefined
Ejaaz: they're affected from it and they can pick and choose the winners. (21:49):
undefined
Ejaaz: There's also the argument that like all this AI CapEx investment is kind of (21:51):
undefined
Ejaaz: a race to the bottom because whoever gets to AGI first is gonna absolutely disrupt (21:54):
undefined
Ejaaz: and destroy any other competitor. (21:59):
undefined
Ejaaz: An argument that you could potentially give. The other side of it is, (22:01):
undefined
Ejaaz: Josh, I really like your point around the personalization and localization and (22:05):
undefined
Ejaaz: the importance of privacy of having your model on your own device. (22:09):
undefined
Ejaaz: Who cares if there's a general query that can be answered by some research professor versus an AI model? (22:13):
undefined
Ejaaz: I want it to be able to read my text and understand the next thing I'm going to text. (22:19):
undefined
Ejaaz: I want it to be able to download the apps ahead of me even knowing about the (22:23):
undefined
Ejaaz: apps and do the things for me. That's only going to work on personalized devices. (22:26):
undefined
Ejaaz: And as you said, Apple has the distribution. (22:30):
undefined
Josh: Yeah, the use cases for local inference are kind of remarkable when you think (22:33):
undefined
Josh: about having access to all of your texts or emails. (22:36):
undefined
Josh: There's a second thing to this also, which is cost. (22:39):
undefined
Josh: If you remember in a previous episode, we covered Grok Nano, (22:42):
undefined
Josh: which was this very lightweight, very easy to use, very low cost model, (22:45):
undefined
Josh: and how within days of it being released, it went to the very top of all the (22:50):
undefined
Josh: charts in terms of usage. (22:54):
undefined
Josh: Every developer in the world wanted to use it because it was fast and it was cheap. (22:55):
undefined
Josh: And one of the trends that we see as AI develops over time is this race to the bottom. (22:59):
undefined
Josh: The cost per token, the race to zero is fairly aggressive. (23:04):
undefined
Josh: And what Apple's done is they've kind of cheated and they've gone from the starting (23:07):
undefined
Josh: line directly to the finish line without doing the work in between because inference (23:11):
undefined
Josh: on a local device that has an M-series chip or an A-series chip that Apple has is completely free. (23:15):
undefined
Josh: It costs $0 to run a query on these local devices. (23:20):
undefined
Josh: And there's billions of them that are capable of doing this. (23:24):
undefined
Josh: So if you're a developer who wants to build a great experience for a customer, (23:26):
undefined
Josh: there is absolutely no reason why you shouldn't be building for Apple devices (23:30):
undefined
Josh: because the inference costs are free. (23:33):
undefined
Josh: And a lot of these developers who are building applications for the world of (23:36):
undefined
Josh: AI, they're bringing these servers. They're constantly looking for the next (23:38):
undefined
Josh: cheapest one because they want to keep the cost down. (23:41):
undefined
Josh: But if you just throw an app in the App Store that runs on these local models (23:44):
undefined
Josh: that Apple will hopefully produce in partnership with Google, (23:47):
undefined
Josh: You suddenly tap into this gigantic context window, this gigantic bay of memory, (23:50):
undefined
Josh: and it costs you absolutely zero dollars to do. (23:54):
undefined
Josh: And I think market forces will kind of come into play at some point here where (23:57):
undefined
Josh: just the natural tendency to trend towards lower prices will allow a lot more (24:01):
undefined
Josh: developer adoption and to create the experiences on top. (24:05):
undefined
Josh: So while Apple completely and utterly failed at AI version one with Siri, (24:08):
undefined
Josh: it is an abomination. It is the worst AI product of any major company by far. (24:12):
undefined
Josh: And I don't want to get that twisted. they may have accidentally (24:16):
undefined
Josh: just by the fact of vertically integrating these chips stumbled into the best (24:20):
undefined
Josh: case scenario where everybody's spending a bunch of money everyone is trying (24:25):
undefined
Josh: to build the ai first get the user second apple has all the users and they have (24:29):
undefined
Josh: all the data they just need the ai to plug into it and google could be that person to do so yeah (24:34):
undefined
Ejaaz: It's super smart i think the market is finally recognizing that if i were to (24:39):
undefined
Ejaaz: boil all of this down to its core i would say that Apple is actually doing what (24:42):
undefined
Ejaaz: they've always done the best, (24:47):
undefined
Ejaaz: which is create the best consumer experience and products. (24:50):
undefined
Ejaaz: And so rather than jump into the infrastructure wars and the infrastructure (24:54):
undefined
Ejaaz: race in AI with GPUs, they let everyone else do it. (24:57):
undefined
Ejaaz: They cherry pick the best, and then they integrate it in the best, (25:01):
undefined
Ejaaz: most artful way for their own products and services. (25:05):
undefined
Ejaaz: And that's why they have the addicted audience that they have, right? (25:08):
undefined
Ejaaz: I buy iPhones practically every single well, you're having MacBooks at this point. (25:13):
undefined
Ejaaz: I'm addicted to the Apple brand. And there's a reason behind that because it (25:17):
undefined
Ejaaz: has the best all-in consumer experience. (25:20):
undefined
Ejaaz: And that's what Apple knows that they have. And they can afford to basically (25:23):
undefined
Ejaaz: sign these deals and partnerships with the Googles of the world. (25:26):
undefined
Ejaaz: Another thing that I find interesting about this, Josh, is in this case, (25:30):
undefined
Ejaaz: they're paying Google to create their own AI model for them, (25:34):
undefined
Ejaaz: which seems like an L on the balance sheet for Apple. (25:37):
undefined
Ejaaz: But if you play this out, Gemini and Google are definitely going to build their (25:41):
undefined
Ejaaz: own agent, right, which is going to use a bunch of apps and they need to get (25:46):
undefined
Ejaaz: access to billions of consumers. Who has that? (25:51):
undefined
Ejaaz: Apple. I know Google already has it across an array of their different products (25:55):
undefined
Ejaaz: and apps that we explained earlier on. (25:59):
undefined
Ejaaz: But Apple has that addictive user experience that only operates at the hardware (26:01):
undefined
Ejaaz: level through the cellular device. (26:06):
undefined
Ejaaz: And if they maintain that lead, that could be a case for Google actually paying (26:07):
undefined
Ejaaz: Apple a hell of a lot more money to get access to this. (26:12):
undefined
Ejaaz: Is that fair? Is that crazy? Or have I got this wrong? (26:15):
undefined
Josh: Yeah, at the edge, it makes sense that the search thing happens (26:18):
undefined
Josh: again where okay apple owns the user experience (26:21):
undefined
Josh: google owns the software stack google winds up paying (26:24):
undefined
Josh: apple for the exposure um like i said before (26:27):
undefined
Josh: a lot of these companies they they have the intelligence they don't have the (26:30):
undefined
Josh: products they don't have the interface they don't have the trust or the users (26:34):
undefined
Josh: apple has that and by licensing technology from gemini they exclude themselves (26:36):
undefined
Josh: from all the capex fluctuations and the crazy amount of spending without a promised (26:42):
undefined
Josh: return in exchange for creating a a (26:45):
undefined
Josh: AI experience that is frankly the only one that I really want. (26:49):
undefined
Josh: The one that has all of my context. It can read all of my texts, all of my emails. (26:52):
undefined
Josh: It could be the personal assistant that no one else can be yet. (26:56):
undefined
Josh: And that's remarkably powerful. (26:59):
undefined
Josh: And if I want to outsource my thinking to an impressive AI, I could do that. (27:01):
undefined
Josh: And sure, Apple could probably ping some sort of API through Google or whoever (27:05):
undefined
Josh: they're partnered with, like OpenAI. (27:08):
undefined
Josh: But for most of my tasks, most of my AI needs, I just want a really helpful assistant. (27:10):
undefined
Josh: And no one can provide that better than an AI on my iPhone. (27:15):
undefined
Josh: And I think that's kind of the bull case in summary is as AI is able to remove (27:18):
undefined
Josh: themselves from CapEx, as they're able to implement local inference on all these (27:22):
undefined
Josh: devices for zero dollars, there's a lot of network effects that will probably (27:26):
undefined
Josh: come swinging their way. (27:30):
undefined
Josh: And maybe it was accidental, maybe it was by design, but the long tail effects (27:31):
undefined
Josh: of Apple's position right now seem fairly optimistic. (27:35):
undefined
Josh: And I think that's probably why it's priced sitting right at an all-time high (27:37):
undefined
Josh: right now, because we're not the only ones that think that. (27:40):
undefined
Josh: And that just about wraps it up. (27:42):
undefined
Josh: Why we think Google and Apple are both undervalued while at all-time high. (27:44):
undefined
Josh: And while perception might not always match public market sentiment and i think (27:47):
undefined
Josh: it's just it's important to continue to evaluate these companies like we said (27:51):
undefined
Josh: on yesterday's episode um these things matter a lot and how ai is implemented (27:53):
undefined
Josh: really matters a lot and it's changing (27:57):
undefined
Josh: Architecture of the world around us. So it's this exciting conversation we're (27:59):
undefined
Josh: going to continue having as we go. (28:02):
undefined
Josh: I just, as always, thank you so much for watching the episode. (28:04):
undefined
Josh: We really appreciate all the support. (28:07):
undefined
Josh: Things have been going well. In fact, so well that 85% of the people who watch (28:08):
undefined
Josh: this episode on YouTube are not subscribed, which is a problem. (28:12):
undefined
Josh: So if you are watching this and you are not subscribed, please go ahead, (28:16):
undefined
Josh: click that subscribe button and also listen to us wherever you find your podcast. (28:19):
undefined
Josh: One place we're struggling to grow is on the RSS feed on Spotify because it's (28:23):
undefined
Josh: kind of tough to get people to go there so if you listen and you want to support (28:27):
undefined
Josh: the show please go there subscribe on spotify subscribe on your rss feed wherever (28:30):
undefined
Josh: your player may be apple podcast even if (28:34):
undefined
Ejaaz: You don't want to listen if you're taking a shower just put it on play put it (28:36):
undefined
Ejaaz: in the other room it helps us out a lot. (28:40):
undefined
Josh: That's what my dad does it's great so yeah we just appreciate the support of (28:43):
undefined
Josh: whatever you can give all the comments it's been amazing we're coming off our (28:46):
undefined
Josh: best month ever and we continue to keep our foot on the gas thanks to all the (28:49):
undefined
Josh: support so thank you for watching as always we will see you in the next episode and peace (28:52):
undefined
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
My Favorite Murder with Karen Kilgariff and Georgia Hardstark
My Favorite Murder is a true crime comedy podcast hosted by Karen Kilgariff and Georgia Hardstark. Each week, Karen and Georgia share compelling true crimes and hometown stories from friends and listeners. Since MFM launched in January of 2016, Karen and Georgia have shared their lifelong interest in true crime and have covered stories of infamous serial killers like the Night Stalker, mysterious cold cases, captivating cults, incredible survivor stories and important events from history like the Tulsa race massacre of 1921. My Favorite Murder is part of the Exactly Right podcast network that provides a platform for bold, creative voices to bring to life provocative, entertaining and relatable stories for audiences everywhere. The Exactly Right roster of podcasts covers a variety of topics including historic true crime, comedic interviews and news, science, pop culture and more. Podcasts on the network include Buried Bones with Kate Winkler Dawson and Paul Holes, That's Messed Up: An SVU Podcast, This Podcast Will Kill You, Bananas and more.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com