Apples Unified-Memory Advantage – Shimmy Says Ep 49

In the battle for AI at the edge, the $600 Mac Mini might beat the a hundred billion dollars data center. Hey everyone, you know, I'm feeling that kind of Caribbean beach vibe today, but, um, I'm here for Shimmy says, and we're gonna have a great shimmy says, and we're gonna have another good one tomorrow. So stay tuned for that.

But for today's shimmy said, I wanted to talk about something I came across and that I'm living here at Textron. This is the last week or two, and it's about Apple and ai. You know, I saw a chart that I'm gonna explain to you in a little bit, but before I do, I gotta make a confession.

And no, this isn't some Jimmy Swagger kind of thing. I'm not confessing, confessing anything immoral. But you know what, I gotta be honest with you, I was one of those people in the peanut gallery who was yelling that Apple missed the bus on ai.

You know what I I, we said it on Textron gang, I wrote about it. I talked about it. I just thought that Apple missed the party, the AI party, and they were gonna be too late to catch up, minus spending an awful lot of money, buying some up and coming AI company.

They were gonna have to dip into that, you know, legendary hoard of cash that Apple sits on. But you know what, they are not the only ones that were sitting on a, a hoard of cash. The fact of the matter is, all of the hyperscalers, all the tech bros were sitting on hoards of cash, meta, Google, Amazon, Microsoft, they all amassed just wads Oracle.

They, they were just sitting on a ton of cash. So what happened? Well, apple does what Apples always does.

It seems they, they make the le let they, excuse me, they make the rest of us look like we're arguing on the sidewalk while they're upstairs, quietly rearranging the furniture in the building. Now, I'm not saying Apple has this all figured out. I'm not handing out participation or winner trophy trophies.

I'm not even saying Apple did this on purpose. This may be purely by accident, but I am saying something is happening and it, it is such an Apple move that it, it like hurts. And so the proof's in this chart that I told you about that's been floating around, I saw it posted up on X by a guy named Josh Gale.

Shout out to him. And when I saw it, I, I both laughed and my eyes bugged out. The chart is about CapEx, how much money companies have spent on capital expenditure expenditures, primarily data centers, you know, servers, infrastructure, all the heavy metal that we are investing in this AI revolution.

And when we look at the hyperscale, as you see Amazon, Microsoft, Google Meta, as I mentioned, they are in, make no mistake, a spending's arm race. They're not investing, they're not prioritizing, they're not leaning in. I'm talking about they opened up their huge piggy banks and spent every do dollar in there.

Then they borrowed money from their neighbor's piggy banks. And then they sent both piggy banks on fire for warmth. They're spending collectively like a hundred billion dollars a quarter.

That's $400 billion a year, almost a half a billion dollars, uh, a year into data centers and AI infrastructure. And then there's apple, Apple's way at the bottom. You, so these lines all go up like hockey sticks.

Apple's line goes down. They actually spent 19% last this year on infrastructure. Their CapEx is down.

Like, nah, we're good. We're not, we're not playing that game home. Don't play that game.

Let me just pause right here for a second because people are going to twist this. I'm not saying Apple isn't spending money. Of course they're spending money and they can afford to spend money.

They got a lot of money. They, and they do spend a lot of money. Matter of fact, they have more money than a lot of countries you may know.

But while the rest of big, big tech has been out there building cathedrals to AI monuments to AI factories, apple is like the guy at the poker table who, who's barely putting any chips in, but somehow he keeps taking the pot on every hand. And that's why I'm calling today's Shimmy says Apple crazy like a fox because, because either Apple is the luckiest tech giant in the history of capitalism or the luckiest player in the history of capitalism, or maybe they're playing 3D chess while everyone else is playing checkers. I don't know.

Well, except for Jensen Wong, of course, right? Yeah. You can't say anything about someone who built a $5 trillion market cap.

So he could be playing whatever game he wants and charging admission. It's all good. But the rest of the big tech guys, they're spending sailor, they're spending like sailors drunk on shore leave.

You know, that's an old saying. I don't know the, the sailors still spend money on shore leave, like drunken sailors. The whole concept of a drunken sailor, I don't know.

But you get the point. They're spending a lot of money. They're spending their money, they're borrowing money, they're spending their neighbor's money, they're spending their children's inheritance.

These companies are investing everything they have while Apple sits back. So obviously Apple didn't build the AI data center, but it looks like they're building the AI machines. And this is what makes this story so frigging man, the hyperscale, while the hyperscalers were shoveling money into giant centralized infrastructure, apple either on purpose or not, seems to have backed into becoming the AI infrastructure.

Not that they built big data centers without realizing it. It's that they shipped computers that everyone is actually using to run AI today. Let me tell you what I mean by that.

Mac Minis, good luck trying to get one. They're selling out Mac Studios, it's like a six week backlog. And I know this isn't just a rumor guys, 'cause it's happening to us.

We ordered several Mac Minis this week alone, just this week. Why did we order Mac Minis? Well, turns out everyone wants to run agents for this agent ai.

What are you gonna run your agent on? You're gonna run it in the cloud. Yeah, probably not.

You wanna run it locally on the edge or what are you gonna run it on locally on the edge? You gotta set up something. Well, it turns out if you want a good practical little machine that don't cost a lot of money, that can just sit there, not use a lot of power, run quietly and act as a node to run agents or local inference workloads, the Mac Mini is really becoming the default answer.

And now that we can't get them as quickly, my guys are out here scavenging any Mac with an M1 or better in the building. Congratulations, you just got promoted. You're a server.

Now imagine saying that five years ago, right? We're turning the MacBooks into servers, but people would've looked at you, you know, like you were ordering sushi at a steakhouse. I've had good sushi at steakhouses.

But anyway, here we are. And the reason isn't Mark magic, it's architecture. Apple, silicon, whether knowingly or not, is a great little AI machine.

It has unified memory. CPU and GPU share the same pool. They've got those neuro nets we've got.

These machines can handle model workloads in a way that doesn't map cleanly to the old GPUV. Ram is the only thing that matters. Worldview a big unified memory machine can run surprisingly large models locally.

And this brings me to the story that makes me stop smile, because it's the exact kind of plot twist you'd expect with Apple no matter what. They just seem to, you know, they're the lucky kid who always got the pretty girl scored the touchdown, they seemed to get it again and again. They look like they were late to the party and they end up being early on a whole different lane.

It's almost like Forst Gump kind of, you know, a lifestyle there. And here's the important thing too. We, you know, for the last couple months we started talking about it's not about training, it's about inference and all these companies were trying to get chips for inference.

Broadcom, there's a huge inference. Uh, de Tanium, the Google inference chips, all of these inference chips. Well it may just turn out Apple has the best inference chip and inference computing platform out there for us.

So why does this kind of stuff happen to Apple though? It they, I you can't believe in luck. Let's ask the question though.

'cause it makes the rest of the industry look like they panicked here. And how does that happen? How does that company on paper that spent the least amount on AI infrastructure end up in a position where their hardware is becoming the preferred way to run AI locally?

It's just Apple, man. It, it keeps coming back to Apple. There's gotta be something.

I wish I knew what it was. I wouldn't be here doing this. I'd be on an island, you know, laying in the sun while my MacBook listening on my AirPods probably.

Uh, actually I do live on an island and I could go to the beach, but it's not a great day. But you know what I'm saying. The point is, apple time and time again, seems to be in the right place at the right time to maximize the opportunity.

They don't, you know what, back in the day they didn't chase enterprises when everyone was chasing enterprises, yet. You go into any enterprise, you'll see them full of max anyway. They didn't chase data centers when everyone was spending their mother's money on data centers the way the hyperscalers did.

Now it looks like they may end up owning the devices that sit at the edge where a lot of the AI's gonna work on. They don't always win by being the first to the prize, but they win by being the one who finishes the race first. They win by being the one that ships the everywhere version.

The version that's in your hand, in your desk, in your backpack. That's Apple secret superpower. That's their secret superpower.

Now look, this game's not over and the hyperscalers aren't all wrong. We're going to need these big data centers though. Construction on them, slowing down.

Everyone's yelling. Not in my backyard. But let me be fair for a minute 'cause I could hear the comments shimmy.

You don't train GPT class models on Mac mini minis, as I said before. No you don't. But you do run inference on them.

If you are going to be still training the frontier models, you need those master GPUs. You need Nvidia, you need a data center budget that makes your CFO cry at night. The hyperscalers are spending because they have to, the big models that they wanna run require that kind of horsepower.

But training's only one part of the AI story. And in many ways it may be moving into the background, running models, inference, serving models, building agents, deploying as agent assistance, keeping things private, not up in the cloud, keeping offline prototyping without sending your data to uh, someone's cloud. These are all a different story than just training.

And this is the where the growth is right now. And I mean growth, it's growing fast. If, if Techstrong hears any indication, it's exploding.

And that's the part of the story that Apple seems to have gotten right. It looks like their devices are the place and things to use there. It's not cloud infrastructure, it's edge infrastructure, consumer distributed infrastructure.

Who knew? Call it whatever you want, but I'm telling you it's real. You know, there's a line from that Expost I mentioned earlier that nails it that believe it or not, the company spending the least on AI infrastructure accidentally became the AI infrastructure.

And that's the story guys. That's the story right there. While everyone else is building gigantic centralized systems, apple built an army, not an army of GPUs in one place.

An army of powerful machines everywhere, just kinda like Buzz and Woody looking out Apples everywhere. And what's wild is Apple didn't even build them for ai, or at least I don't think they did. And they haven't admitted that they built, they did.

They built them because Apple likes to build walled garden controlled things, performance per wat, battery life, tight integration. They wanted to own their own silicon. And now all those decisions happen to line up perfectly with the moment in time where we are for what local AI needs.

Like I said, crazy like a fox because if it's intentional, it's genius. If it wasn't intentional, it's still genius because it came from Apple doing what Apple does. Now, as I said again earlier, this, this don't mean it's over the platform more.

We're still gonna have one. Let's not get too carried away. It can still cost them that they haven't invested in what these hyperscalers has have.

Because the AI war isn't just about the hardware, it's about the platforms, it's about the ecosystems, it's about model availability. It's about distribution of software and services that people build their workflows around. If the future becomes, everything runs in the cloud and the device is just a thin client, then Apple being the local AI machine won't matter as much.

But if the future becomes one or two giant model providers own that layer and everyone else rents intelligence. Apple's position is still less clear there. But I don't know if that's the way it is.

I think we're gonna see AI on the edge, we're gonna see AI personally on, on the person, and Apple can still be the best hardware. But if one of those other scenarios come out, they can still be the best hardware and lose the platform. It's possible, but it's Apple, I wouldn't bet against them.

'cause as the world move moves towards a hybrid future where some things will run centralized, a lot of things will run local, private, near the user. Apple got the central lane, the center lane guys. And that's very much my part where I think things are going.

So what's the punchline here? Let me start where I, or in the end where I started, I was wrong when I said Apple missed the AI bus. Not the first time I've been wrong, not the last time they have missed the loudest bus.

The bus with the flashing lights and the smoke and the guy hanging out the window screaming. We're doing a AI or a GI. But Apple might've built something better than the bus.

They might've built the road. The cars, they put the engine into a billion plus pockets. And while everyone else is spending a hundred billion dollars on data centers, apple's quietly letting the world turn on $599 machines into agent servers.

It's crazy like a fox. You know, I looked for a black turtleneck to wear for this one today, but I don't have one. But I know somewhere Steve Jobs has gotta be smiling about this.

Take care at Shimmy for this week. We'll see you next week. Says.