Apple Won the AI Race

What I think I hear you saying is there's actually two different races.

Yeah.

In race number one, the competition is so far ahead, Apple's kind of already took the L. But in race number two, the finish line is way out on the horizon. And Apple does have a bit of a head start because they still make the iPhone.

So that's the smartest moment in Marquez Brownley's entire video. Two races. There's the model race and the hardware. They built the exact right frame. And then the whole conversation zooms in on the model race, which sounds reasonable until you look at what winning that race is actually worth. June 9th, Enthropic drops Fable 5, best model on the planet. The Independent Intelligence Index scores it at a 60. Clear top of the board. 30 days later, OpenAI ships Soul. Soul scores a 59. One point back at a third of the cost per task. 30 days. That's how long the best lead in the AI race lasted. Nobody's winning that race. Nobody can win that race. And the second you understand why, this whole debate flips upside down. And that new model, the one that closed the gap in 30 days, it got caught cheating.

Let's take a look at Marquez's video.

Yo, I was just thinking about all this AI race stuff and how everything's gone from zero to 100. And then Apple, they've just kind of been scrambling the past two years trying to build Apple intelligence and haven't really done very well yet. I just feel like it's so obvious that they lost.

What? No, they they lost.

No, they they really won.

Wait, sorry. No. Are you Are you trying to say that Apple, the one that announced the new Siri and then didn't deliver on it for two years until they had to start over and build it from the ground up with help from Google, that Apple won the AI race?

See, I I think your definition of the AI race is a little bit too narrow. Like obviously being open AI and having chat GPT is one thing, but making the hardware that everyone runs AI on is a different thing.

Yes. But also, I don't think you can ignore how hard Apple has lagged here. Like I remember when ChachiPT first came out and everybody saw this as like the obvious huge paradigm shift. So if you're a tech company, you got to move fast. So they did. Google starts going crazy. Meta turns the burners on. Bing was insane for a couple of weeks. I remember all this action. And then Apple just nothing. Just nothing from them. They literally refuse to say AI in their events for months.

I know, but think about it. It's Apple, right? Their thing has never been being on the bleeding edge of tech. They always sit back, let the early adopters figure out what the pros and cons and the bugs are, and then they'll step in late with their own version, taking it to the next level for their ecosystem, right?

Oh. Oh. Oh.

Yeah. I mean, that's the standard Apple defense, right? Apple's never first. They sit back, they let everyone else work out the bugs, and then they step in late and do it better. All those rumors of the foldable iPhone years after Samsung had a foldable Galaxy, the history supports them, right? The iPod wasn't the first MP3 player. The iPhone wasn't the first smartphone. There's so many examples of this, but listen to what that argument actually is as a personality trait, right? Apple is just like that. He's describing a habit, and a habit only tells you what Apple will do. It doesn't tell you whether it actually works. So, here's when showing up late actually works. When being early buys you nothing that lasts. So, think about it, right? If somebody in the AI race could build a lead and keep that lead, waiting would be suicide. Apple would be locked out of the future right now. So the real question is the only question in this entire debate is whether anyone in AI can build a lead that holds. All right, let's keep watching.

That's that's what Apple intelligence is. Apple waiting for the tech to get good enough for the first 100 million users of chat GPT to work the bugs out. Like that can't be right. But okay, obviously it's past the early adopter phase, but not every technology has to be adopted like some kind of race. Like obviously tech companies are trained to see emerging stuff as a threat. But this AI stuff was never a threat to Apple's core business. Like they're still a hardware company. Like you said it yourself. I've watched your videos. They are the iPhone company and they sell the iPhone. Okay. But you would expect still the company with trillions of dollars to spend would still see chat GPT spin up and then Gemini and Claude and Meta and all this other stuff and would think, okay, number one, this is here to stay and number two, we should build a version of our own that's cohesive with our ecosystem. But Apple Intelligence has been kind of a joke so far. Now, I do have an advantage over Marquez here because the most recent version of iOS that has the new Siri, the new Apple Intelligence in it, uh, just released for demo and it's getting really positive reviews, even though, as most of us know, it's built on top of Gemini, not Apple's own Frontier model. So, keep that in mind. I'm not obviously dinging him for that. Um, the the beta came out after he published this video.

Hot take I don't think Apple Intelligence has to be that good.

What?

Yeah. No, I I basically think it just has to be good enough for investors to believe Apple's competitive or just doing something with AI, but it doesn't have to be amazing for Apple to still be winning cuz you're still buying an iPhone.

Okay, but what if I'm not buying an iPhone? Like if I'm if I'm someone who really cares about AI, then there's so many other phones that I would buy before an iPhone, right? I would buy a Google phone or a Samsung phone or a Xiaomi phone. They all have the AI assistants and the writing tools and the I don't know, the photo editing, all this AI stuff that's way better than what Apple's been making.

Yeah, but dude, everything you just mentioned is such a gimmick. Like, no one's actually buying a new phone just to use that stuff, really. So, Apple Intelligence doesn't need to be that amazing because you're still going to buy an iPhone and all of those AI features that get built into these existing phones are basically gimmicks. No one actually buys a new phone to use all of these AI features.

Okay, so let's think about what Apple lost the AI race has to mean. Either it's a business claim, meaning Apple sells fewer iPhones because their AI is worse, or it's an image claim, Apple just looks bad because they don't have a Frontier model. And to their credit, they just settled on the business claim themselves. No one is switching phones over AI features. So all that's left of the Apple lost is vibes about Apple's reputation. And here's the evidence right from the other side of the fence. Gemini is wired into Android deeper than anything Apple has ever shipped. Chat GPT sits right there on the iPhone on Open AI's app. And the switching still isn't happening in either direction. The models are close enough that even deep integration can't beat the pain of switching phones, switching to a new phone ecosystem. So, the same pattern shows up twice. Once on Android, once on iPhone, that closeness between the models, that's the thread I keep pulling on. And it really goes even deeper than this conversation has time to get into.

It almost feels like Apple's just playing along just to, you know, have some stuff that's comparable. Oh, yeah. Here's some writing tools. Here's some photo editing with AI and and then of course the Siri improvements, you know, Genoji and Genoji of course. But here, tell me if you agree with this. All the fastest, most secure generative AI stuff happens on device, right?

Yeah.

And then the bigger, more capable models, that's what you go to the cloud for. Exactly. So, as these ondevice models get better and better, you go to the cloud less and less, right?

Yeah.

So, in the future someday, you will basically never have to go to the cloud. You'll do all your stuff on device. And who's going to make that device? Apple.

Maybe. Honestly, don't think maybe. I think definitely. Like in this future where all these early adopters finally get their local ondevice models doing tasks for them and running everything locally with tons of memory on these machines, Apple's going to be there to sell those machines. That's where they fit into this AI puzzle. And just because they don't do the software doesn't really mean they're losing. Like they don't do search engines either. And they sell you the device that should do the googling on. So, yeah. Why do you think Nvidia wants to make RTX Spark a thing so bad? They want a piece of that hardware pie, too. Especially outside the US.

Okay. So, what I what I think I hear you saying is there's actually two different races.

Yeah.

In race number one, the competition is so far ahead, Apple's kind of already took the L, but they can't look like they're losing. So, they have eventually cobbled together enough of an AI suite to not be embarrassing anymore. But I think it's pretty clear that in a world where company A pays company B a billion dollars a year, company B already won.

Okay, two races. Race one, the model race. Apple already took the L. Hard to argue that one. They clearly don't have a frontier model. They say they're working on it, though. Race two, the hardware race. Long horizon, Apple has the head start. And then he says this line, right? Company A pays company B a billion dollars a year. So company B already won. But this is where the question from earlier gets answered because race one, the race that Apple supposedly lost. It doesn't have a trophy. June 9th, Enthropic released Fable 5, best model on the planet. Uh billions and billions of dollars in development costs to build this model. They held that top spot for 30 days until OpenAI ships sold. Basically a statistical tie. But Soul is a third the cost per task. I run both of these models every single day in my business. soul burns way fewer tokens getting to the same answer. So in practice that gap is even bigger than the difference in token price. Okay, so the best lead in artificial intelligence lasted 30 days. Now we can look at a test like vending bench on top of that. That's the benchmark where Frontier models run a simple vending machine business over several months. A human baseline can make $63,000. not like the best human in the world, but like average human performance. The best AI frontier models are currently doing about $8,000 uh or below. They're all clustered together, all the frontier models, but way, way below the human. Now, I broke that research down in another video, but the takeaway for today is this. These models are converging. Everybody keeps improving, and they're improving fast, but everybody is improving together. Every jump with the next model is getting matched within a few weeks. Nobody is running away with the race. And a race where nobody can run away pays nothing to the winner because they're only going to be the winner for a little bit. So now come back to his billion-dollar line. Company A pays company B wins. But as of very recently, we can test that claim. Apple opened the iOS 27 public beta. The headline features the new Siri. For the last 2 years, people have been laughing at Siri, but early reviews of iOS 27, the new version, is calling it a redemption. Testers are saying it finally matches the demo that Apple promised back in 2024. And as we know, under the hood of that Siri lives Google Gemini. Apple didn't build the Frontier model. They saved billions and billions of dollars not building a Frontier model. Instead, they took someone else's model, distilled it down to run on their own hardware, and shipped it toward two and a half billion devices. Google won a supplier contract. Samsung makes the screens for the iPhone. Samsung, one of Apple's biggest competitors in phones. And no one says that Samsung won the phone race because they make the screens. When every part is good enough, a part is just a part. Components don't own the customer. The device does. And one more thing about Soul. That's the model that closed the gap in 30 days. I said it got caught cheating. Here's what happened. MER, the independent safety organization that evaluates these models, it found that Soul gained its software engineering evaluation at the highest rate they've ever recorded. The benchmark it's winning is the benchmark that it's basically cheating on. So the scoreboard is of course compromised. But the prize for this race just doesn't exist. Now if the models can't decide this fight, something else will. And let me tell you a story from the '90s. I'll give away my age here a little bit. I have a story from the '90s that's going to tell you exactly what. But I want to see where the two Brownley's land.

Fair enough. But in race number two, the finish line is way out on the horizon and Apple does have a bit of a head start because they still make the iPhone and Apple is going to protect that iPhone's position with everything they've got. Now, they could still win that race in a way, but also there's nothing preventing Open AI from coming along and making a phone and taking everything from Apple, the iPhone company. There are so many rumors of them working on something like that already.

Yeah. Okay. So, Apple's gonna enter a new product category and they're going to do it the exact same way they always do, which is by making it work the best with the iPhone. And actually, that's what they're doing with the Have you tried the beta, the new Apple Intelligence stuff? Like Siri pulls from iMessage and Apple Calendar.

Oh, I didn't realize that. Uh I I guess it makes sense, Mark. He's got early access to this. So, he was aware of how much better iOS 27 is. and photos in a way that the chat GPT app on the iPhone never could. But Siri doesn't like excel at anything that chat GPT or Gemini doesn't already do. Like we're not coding apps or anything with Siri. It it doesn't even work in the background. So you can't leave it and come back later for large requests like you can in the others. It also doesn't remember anything about you without making notes in the iPhone's notes app. So it just has way less of that personal context. Like that's just basic stuff you can already do today with the Chat GPT app on the iPhone.

Yeah. Okay. So, the real question is, do you think an AI company is going to come along and make some new hardware that's so good that I can get people to switch from the iPhone or do you believe Apple could make their AI good enough to keep people on the iPhone?

So, I won, right? All right. So, the the close there was almost exactly the right question. Completely agree. Can an AI company build hardware good enough to pull people off of the iPhone, or can Apple make its AI good enough to keep them? So, here's the story that I've been thinking about while I've been watching the Brownley's uh argue. Okay, back to the '90s. Uh Sega versus Nintendo. These are two companies at war to be the best at the exact same thing. Trading the lead back and forth for years and neither one ever pulling far enough ahead to end this race. That is OpenAI versus Enthropic versus Google right now versus X, you know, whatever. Trading a one-point lead every 30 days. But here's how the console war actually ended. Sega didn't win and Nintendo didn't win. Sony walked in from the side with the PlayStation and essentially took the entire market. And that's the real fear behind the Open AI phone, right? Open AAI is Sony here. Apple is about to get PlayStation. Except we have to look at why Sony won. The PlayStation was a generational leap over the Nintendo and the Sega. It was so much better that people switched their whole setup for it. And the data says no one in AI gets to keep a leap like that. Every jump is matched in a month or so. No runaway model means no PlayStation can exist in this race. It means everyone stays Sega and Nintendo forever trading a lead that no one can keep. And there's a second condition Sony needed. The PlayStation was exclusive, right? You couldn't get it anywhere else. You had to buy the PlayStation to have a PlayStation. But if we look at AI today, ChatgPT, Gemini Claude, every single one of them is sitting on the iPhone as an app free to download. These models are Apple's guests. Every time one of them gets better, the iPhone itself gets more useful. And Apple doesn't spend a dime doing it. So for an OpenAI phone to actually hurt Apple, OpenAI has to do two things at once. It has to build a model so much better that people abandon their photos, their messages, and 10 years of muscle memory for this new OpenAI phone. And the data says that they're not going to be able to build that lead. On top of that, right, they then would have to pull chat GPT off of the iPhone so that it's exclusively on their own phone, which means walking away from one of the biggest distribution channels they have to bet everything on hardware they've never shipped. An OpenAI phone with a good enough model and no exclusivity is just another Android trying to out hardware Apple. And companies have been dying on that hill for 15 years now. before the Apple fans get too comfortable. This doesn't mean Apple's safe. My argument doesn't crown Apple. It moves the fight. If the models are all good enough, then AI turns into a component and components flow through whoever owns the customer. Okay? So, that fight is real. It's open and Apple is strong there, but not untouchable. So, don't worry about did Apple lose the AI race. The real question is whether anyone can dislodge Apple now that AI has collapsed into a part that anybody can buy. Okay, that's the question their conversation walks right up to and it's the only one that matters. Now, everything I just showed you sits on top of one claim that no AI company can build a lead that lasts. Now, yes, the models haven't stopped improving. They're improving faster and faster than ever, but every jump gets matched and the long horizon autonomous work is exactly where the whole pack is still stuck together. Now, if you want the research on why those fully autonomous agents Silicon Valley keeps promising haven't actually shown up yet, that's this next video right here. Go check it out.