Limitless: An AI Podcast

Josh:
If you're a fan of Apple products, this is an incredible week.

Josh:
And even if you're not, chances are something that was just announced in the

Josh:
last three days is for you.

Josh:
Apple just announced over five products this week. But there is a headline within

Josh:
this that everyone's missing.

Josh:
I think a lot of the headlines point to the fact that prices now start at $600,

Josh:
that the specs are now really high, that you can run OpenClaw on a laptop from

Josh:
anywhere in the world at a pretty high rate of performance with models that

Josh:
actually run locally on the device. These are all very cool things.

Josh:
But one of the things that people are missing is the AI angle to all of these

Josh:
releases and how Apple is quietly becoming one of the biggest players in the

Josh:
AI space, even though they haven't actually spent any money on building AI models

Josh:
or scaling their infrastructure.

Ejaaz:
This is probably the most bullish I've been on Apple in the AI race.

Ejaaz:
And the funniest part is that they've made no mention of an AI device,

Ejaaz:
but that's exactly what they released. You mentioned five devices.

Ejaaz:
We've got, what's it, a MacBook Pro. There's a new laptop and a specialized chip.

Ejaaz:
I've got my turtleneck on today, Josh, in honor of Steve Jobs.

Josh:
I'm going to be Steve Jobs, Sarah.

Ejaaz:
For those of you who are new to our Apple episodes, we have the number one fan

Ejaaz:
of Apple on the show. That is Josh. It is his Super Bowl.

Josh:
I'm fired up.

Ejaaz:
Today, we're going to get into the weeds about why Apple just released the top new AI device.

Josh:
Okay, so let's first get into what they actually released.

Josh:
Because there's quite a few things they released the macbook pro they released

Josh:
a macbook air a brand new macbook named the macbook neo which starts at a surprising

Josh:
price point of 600 this is important remember that they released the studio

Josh:
display and studio display xdr pause right there because these studio displays

Josh:
are awesome the previous xdr display used to cost 6 000 this

Josh:
way better specs across the board so i will be trying to purchase

Josh:
one of those bad boys and then finally they released the ipad

Josh:
air and the iphone 17e those two devices

Josh:
are also priced at 600 so for the first time there are three entry-level apple

Josh:
devices priced at 600 bucks and this is important and this is noteworthy what's

Josh:
also important and noteworthy that wasn't mentioned as much is the chip architecture

Josh:
that lives within these devices particularly the new MacBook Pro and the MacBook Air.

Josh:
Now, Apple created their vertically integrated silicon back with the M1 chip,

Josh:
the M series chips, and they are now on the fifth iteration.

Josh:
But this fifth iteration is pretty amazing. And I think that's the product that

Josh:
a lot of people are sleeping on today is not only now do they have the MacBook Neo,

Josh:
the iPhone 17e and the iPad Air that are capable of running Apple intelligence

Josh:
locally for $600, but now they have these chips that are capable of running

Josh:
actually large language models locally on a MacBook. And this is the first time ever.

Ejaaz:
I think this is the most excited I've been about what's inside an Apple product

Ejaaz:
versus what's actually outside.

Ejaaz:
I'm usually a display guy. I'm usually an iOS app guy. I'm like,

Ejaaz:
wow, this experience is amazing.

Ejaaz:
But these chips are actually insane. So let me give you the rundown of the headlines.

Ejaaz:
The AI compute processing power

Ejaaz:
of an M5 chip is around 4x larger than the previous generation of M4.

Ejaaz:
It's 8x larger than the M1, so the first in this entire series.

Ejaaz:
Now you can do a bunch of AI prompts, toolings, apps, and system integrations

Ejaaz:
on your laptop, and it just feels buttery smooth.

Ejaaz:
In fact, it's just super quick. And the reason why this makes me really excited

Ejaaz:
is now you can conceivably host and run AI models on your own local device.

Ejaaz:
That means you can give private access to private data without handing that

Ejaaz:
over to the likes of OpenAI or Anthropic.

Ejaaz:
And you can create a more personalized AI experience without having to hand

Ejaaz:
over all that private data.

Ejaaz:
But there's a really unique architecture around how these chips are made.

Josh:
Before we get into the novel architecture, I just want to double down on something

Josh:
you said, Ejaz, which is a testament to how fast this is relative to previous hardware.

Josh:
One of the fun facts that I love is that previously, when Meta was releasing

Josh:
the LLAMA model, the $70 billion model, that required something like $40,000

Josh:
worth of GPU clusters to run 18 months ago.

Josh:
And now you can run that on this new M5 chip. And that brings us back to the

Josh:
novel breakthrough that enables this to happen, which is the post that we're seeing on screen.

Josh:
They basically took what other companies were calling

Josh:
the chiplet architecture and built their own version of it where you take the

Josh:
cpu and you take the gpu two of these things are both very important to processing

Josh:
ai and they fuse them together into a singular chip and what's interesting about

Josh:
this is the cpu part of the chip is the same on every single version of the ship there's the m5

Josh:
chip there's the m5 pro and there's the m5 max all of

Josh:
those have the exact same cpu the only difference is the amount of

Josh:
gpus that they bolt on so the pro gets 20

Josh:
cores the max gets 40 cores and you could think

Josh:
of it like these lego blocks for apple silicon and this

Josh:
is noteworthy because you can scale this a really

Josh:
long way what we know about ai models in general is that

Josh:
gpus are how you scale these things and the cpu is

Josh:
kind of used as the orchestration layer so this allows

Josh:
these new chips to be modular in the sense that they can just kind of

Josh:
stack gpus more and more and more and i assume this is the architecture that

Josh:
we're going to see with the ultra chip that's probably coming out later this

Josh:
year that's going to be able to run some serious ai models locally on this device

Josh:
and it's a really novel way of architecting these m-series chips that apple's

Josh:
kind of doubled down on and i think

Josh:
we're going to see some really amazing improvements from it what i like

Ejaaz:
About the modular approach is um it doesn't seem to come at the cost of the

Ejaaz:
size of the device like these things are still getting smaller and sleeker and

Ejaaz:
thinner every single year and generation.

Ejaaz:
So that makes me like, you know, super excited about like how much further we can take these devices.

Ejaaz:
The other thing is I watched a really unique video this week where some random

Ejaaz:
dude hacked into Apple's, I think it was their M4 chip. I don't know if you saw this, Josh.

Josh:
This is so cool.

Ejaaz:
Yeah, they converted it into an AI transformer.

Ejaaz:
So what that transformer was capable of doing was training, inferencing,

Ejaaz:
or fine-tuning an AI model right there on his Apple MacBook.

Ejaaz:
And what he found out was the inference and training costs were 80 times more

Ejaaz:
efficient than an NVIDIA GPU, an A100.

Ejaaz:
Now, that's from some lone person hacking into this. And Apple's obviously,

Ejaaz:
not necessarily seen this example, but they're aware that they have the most bleeding edge chips.

Ejaaz:
There is no consumer tech hardware

Ejaaz:
device that has more premium components making it up than Apple stuff.

Ejaaz:
Like their supply chain is just insane. So I'm really excited about that.

Ejaaz:
The other thing is what this unlocks is, in my opinion, what I'm calling a new

Ejaaz:
era of personalized intelligence.

Ejaaz:
I think one of the main challenges that AI is faced today is that we're relying

Ejaaz:
too much on model labs, which kind of results in a more fractured experience.

Ejaaz:
Like the model doesn't know who we are. It keeps asking us to tell us about ourselves.

Ejaaz:
And with this new chip architecture, you can have a more persistent AI agent

Ejaaz:
that understands who you are, that is more useful, that is there right with

Ejaaz:
you in the weeds as you're doing stuff on the internet or on your computer.

Ejaaz:
And that's really bullish for me.

Josh:
I can't help but imagine what it would look like if Apple decided to really

Josh:
scale the manufacturing production of these chips and turn them on for AI training,

Josh:
similar to what Google did with their TPUs, and kind of have their specific

Josh:
hardware accelerated version of these GPUs.

Josh:
I feel like that would be a huge business opportunity, but clearly they are

Josh:
not taking advantage of this position they're in.

Josh:
Because previously on an episode a few weeks ago, we spoke about CapEx,

Josh:
how much money these companies are spending on scaling AI infrastructure.

Josh:
This includes data centers, this includes GPU, powering the data centers,

Josh:
all of the employees that are required to make this happen.

Josh:
And the numbers were staggeringly large i mean

Josh:
between what is this amazon google microsoft meta

Josh:
we have over 630 billion dollars of spend

Josh:
but apple is only at 1.4 which is

Josh:
actually down 19 year over year you

Josh:
see this little sponge down here it's like so sad and depressing and they're

Josh:
spending no money on scaling this and you have to ask like why what is going

Josh:
on here um because clearly they're in a position where they can win if they

Josh:
double down on these things that they're working are working well,

Josh:
but they're just not doing it.

Josh:
And I wonder if you have any takes on this of what you think...

Josh:
Like, what are these other companies doing? And why is Apple not participating?

Ejaaz:
So there's an optimist take on this story. And then there's a pessimist take.

Ejaaz:
The pessimist take is Apple was asleep at the wheel and they were not focused on AI.

Ejaaz:
They completely missed that rush. And they fell behind creating one of the leading

Ejaaz:
intelligence models when they are the most valuable company.

Ejaaz:
So Google, Microsoft, Amazon, and all these companies that you're seeing on

Ejaaz:
the screen here got way, way ahead.

Ejaaz:
Now, the optimist's take is this was all planned because Apple's decision was

Ejaaz:
never to partake in the AI model race.

Ejaaz:
Apple's plan was to own the distribution and operating system layer of AI,

Ejaaz:
which is, hint, hint, what they did with cell phones and the App Store and iOS,

Ejaaz:
and they're doing exactly the same thing on AI.

Ejaaz:
So you could actually look at it as Apple was so smart, not wasting hundreds

Ejaaz:
of billions of dollars of their hard-earned cash, and instead pays Google a

Ejaaz:
billion dollars to rent Gemini and then build an ecosystem right on top of it.

Ejaaz:
It's kind of genius if you think about it.

Josh:
This is like the ghost of Steve Jobs hand, like looking over the company.

Josh:
And actually, if you scroll up on this post a little bit, it's kind of me making

Josh:
fun of Apple and how they've kind of accidentally stumbled upon this miracle.

Josh:
And I would believe the optimist case, if in fact they didn't totally fumble WWDC two years ago,

Josh:
there was a very clear intention to deploy Apple intelligence throughout the

Josh:
suite of hardware and to place themselves into this AI race,

Josh:
it just failed completely and catastrophically.

Josh:
And had that not have happened, I think I could have believed this optimist

Josh:
take where they really are just being slow and calculated. But I think this was an accident.

Josh:
Just so happened to create the best hardware in the world.

Josh:
And it just so happens that all the AI models need to run on hardware just like this.

Josh:
It's kind of like NVIDIA. Like NVIDIA accidentally became the most important

Josh:
company in the world. And it required a lot of execution along the way.

Josh:
And they deserve every bit of that. But they were in a unique position to do so.

Josh:
And I haven't seen any signs of Apple doubling down to do so on the software side, at least.

Josh:
It's only been on the hardware side. And it's only because this has been the

Josh:
trajectory since 2021 when they first launched the M1 chip. But it is interesting.

Josh:
I mean, if we look at this chart down at the bottom of this post here,

Josh:
it shows the increase in CapEx from everyone is going straight vertical and Apple spending none.

Josh:
And yet the Mac minis are sold out everywhere.

Josh:
You cannot buy one because everyone's running OpenClaw on it.

Josh:
The Mac studios are running local models on everybody's machines.

Josh:
The new MacBook Pros are incredible.

Josh:
They're going to be running models. I mean, there's just everything is sold out.

Josh:
Everything is backlogged. They can't make enough hardware to support this.

Josh:
And something is happening here.

Josh:
The market forces are at play and they are saying, Apple, you are making great

Josh:
hardware. Please do more of this.

Ejaaz:
I was joking with some friends the other day that the only real threat to NVIDIA's

Ejaaz:
hardware mode is Apple with their Mac minis and with their laptops because they're the second largest,

Ejaaz:
most valuable company that comes

Ejaaz:
behind them for a reason because they're selling out all their hardware.

Ejaaz:
People can't get enough of it, but consumers in particular use it to run their AI models.

Ejaaz:
On the software side of things, listen, it's not clear just yet,

Ejaaz:
but I do think Apple gets ahead for two main reasons.

Ejaaz:
Number one, they have like the largest distribution ever. I believe it's 2.5

Ejaaz:
to 3 billion active Apple devices currently in the world right now. It's insane, right?

Ejaaz:
So if they wanted to, they could switch on bleeding edge AI via Google's Gemini

Ejaaz:
or their own fine-tuned version of that model to 2.5 to 3 billion people tomorrow, right?

Ejaaz:
So they instantly become the most or the largest consumer moat for AI immediately.

Ejaaz:
But they're taking their time.

Ejaaz:
I believe they're building something much more curated and better than what we have today.

Ejaaz:
Now, the pessimists will say, oh, they're slow. They've been slacking.

Ejaaz:
And I would probably agree with you. But hey, they're the second most valuable

Ejaaz:
company in the world. They can take that time to wait and build something interesting.

Ejaaz:
That being said, all these releases are super cool, Josh.

Ejaaz:
But there's one thing that's nagging me at the back of my head,

Ejaaz:
which is Siri AI has been delayed again.

Ejaaz:
I do not know when I'm going to get this, but it's already been delayed,

Ejaaz:
what, a year and a half at this point.

Ejaaz:
This is, of course, Apple's personal AI assistant, which is probably going to

Ejaaz:
be the conduit and the main spokesman for all the Apple software stuff.

Ejaaz:
So until I see that released, I'm not going to believe that it's actually happening.

Josh:
Yeah, I mean, the software part, again, they just fumbled so hard.

Josh:
There's no denying it and there's no signs that they are going to recover.

Josh:
I think the most bullish thing they've done recently in terms of software is

Josh:
just license out their ai to gemini clearly google

Josh:
can do an amazing job and for a billion

Josh:
dollars a year apple is getting access to gemini models and they're going

Josh:
to integrate locally into perhaps not locally but they'll integrate them into

Josh:
all these mobile devices and that takes us to this new weird place where like

Josh:
there is a potential to shift the current market forces based on this distribution

Josh:
that you just mentioned that google that apple has of multiple billions of products

Josh:
already in people's hands that are AI capable.

Josh:
And this is noteworthy. I mean, what we're seeing with OpenClaw in particular,

Josh:
it kind of set the stage for how strong of a preference people have to running

Josh:
these models on actual hardware that they feel that they own.

Josh:
A lot of people bought OpenClaw not because they needed the compute to run the models.

Josh:
That's not true. You could do this on a $5 virtual private server.

Josh:
They used it because it connected with the ecosystem that Apple provides.

Josh:
They used it because it can query through their iMessages and it could send

Josh:
iMessages and it could call and it could FaceTime and it can use the Apple speed of software.

Josh:
And that is a really big deal. And that gets into this AI edge compute bulk case,

Josh:
which is the idea that everyone needs GPT 678910 served from OpenAI's cloud

Josh:
servers might not be true for the majority of the users that actually just want

Josh:
AI to help them like figure out their grocery store order and summarize their emails for them.

Josh:
And there is a limit to the intelligence that the average user

Josh:
Actually need access to. And it would seem as if a lot of these current models

Josh:
have reached that threshold.

Josh:
Not everyone needs to go cure cancer or solve novel physics.

Josh:
And with that understanding, we're at a moment now where Apple's hardware,

Josh:
particularly this new hardware, is able to run all of these models that are

Josh:
capable of these average use cases locally on device.

Josh:
And that's a lot of users that will be using this.

Josh:
And it may actually be like one of the largest bear cases against companies

Josh:
like OpenAI, like Anthropic,

Josh:
who rely so heavily on customers paying money and using the API fees because

Josh:
these local models are becoming so highly intelligent, so capable,

Josh:
and so small that they could just run on an iPhone.

Ejaaz:
Yeah, I actually wrote about this at length in the essay that we just dropped in our Substack.

Ejaaz:
If you're not subscribed, definitely go check that out. It's,

Ejaaz:
in my opinion, a banger, and it goes through everything that Josh just covered.

Ejaaz:
My thinking about models has evolved pretty drastically over the last month

Ejaaz:
in the light of OpenClaw.

Ejaaz:
Because what I initially thought was OpenClaw was just a bunch of open source

Ejaaz:
tech-heavy developers that were just kind of toying around and messing around

Ejaaz:
with something quite dangerous.

Ejaaz:
And then what I actually learned was that the reason why they were doing it

Ejaaz:
was because it led to a better AI experience overall.

Ejaaz:
And when you and I have tried out OpenClaw and we have a bunch of episodes that

Ejaaz:
demonstrate this, we have just had a much better experience.

Ejaaz:
Like the AI actually remembers you but can do so many things for you.

Ejaaz:
And I think that's where people are eventually going to settle,

Ejaaz:
particularly consumers, like, yeah, okay, we can talk to ChatGPT,

Ejaaz:
but I want it to now do stuff for that.

Ejaaz:
It's much harder to do that if your AI provider is open AI with their own servers

Ejaaz:
versus having an AI model locally on your device.

Ejaaz:
So I do think there's a larger trend, which is going to be around edge compute

Ejaaz:
and local AI devices and local AI models running on your phone and on your laptop,

Ejaaz:
which will lead to a more personalized experience. I'm excited to see people take privacy.

Ejaaz:
More seriously, at this point, with OpenClaw, some of the worst examples of

Ejaaz:
it was the agent would steal your credit card info and spend it on some random

Ejaaz:
stuff or would go rogue and burn up all your compute tokens.

Ejaaz:
You have more control and access over that if you go through an Apple device

Ejaaz:
that might kind of give you a semi-private experience that you don't have to

Ejaaz:
expose all that kind of data.

Ejaaz:
If this trend becomes true, then it completely threatens Anthropic and OpenAI's

Ejaaz:
moat, which have relied heavily on subscriptions.

Ejaaz:
Why would you pay $200 a month on a Claude subscription? I'm just playing the antagonist here.

Ejaaz:
If you could get frontier intelligence for a much smaller model that fits on

Ejaaz:
your mobile phone device, that's a QN model that got released this week,

Ejaaz:
and that can work with all your personalized data.

Ejaaz:
Why wouldn't you just do that? It's a no-brainer, and I understand the thesis

Ejaaz:
behind it. And I think that's what Apple's going after.

Josh:
Well, now we have to ask the question, are they capable of doing this and who

Josh:
is going to get them to this place.

Josh:
And to do so, we have to look at the leadership.

Josh:
We have to go to the seed suite first. And that is thanks to Polymarket,

Josh:
who has prediction markets on who is actually going to be responsible for running

Josh:
the ship after Tim Cook leaves.

Josh:
So it's been widely rumored that Tim Cook is going to be stepping down from

Josh:
Apple to as CEO capacity sometime this year. He's been there for a long time.

Josh:
He's had an incredibly successful run, But it seems as if the Apple C-suite

Josh:
is kind of grooming the next person.

Josh:
And according to Polymarket, John Turnness is going to be that guy.

Josh:
And this is exciting because John Turnness is the VP of, I believe,

Josh:
engineering hardware at Apple.

Josh:
He's a hardware guy. He's the person that has helped design,

Josh:
develop, and lead these devices.

Josh:
And Polymarket has him at, what, over 50% chance of running the company.

Josh:
If he does actually become CEO, is there a world in which he can push this company

Josh:
forward in the sense that they can really double down on this edge AI compute thing?

Josh:
They could get these models running on all the devices, maybe.

Josh:
I think that would be a really fun opportunity to see.

Josh:
And we really need a shakeup. Like Apple's been so slow. They've been so boring

Josh:
for so long that this would make a really big difference.

Josh:
And also there's another market that shows the future products that they're

Josh:
planning to launch. And there's one that I think surprises a lot of people,

Josh:
which is a foldable phone before 2027.

Josh:
At an 84% chance. So by September of this year, you will be able to buy a folding

Josh:
iPhone, which seems a little bizarre. But according to Polymarket, this is true.

Josh:
Thank you to Polymarket for supporting and sponsoring this section of the episode.

Josh:
And yeah, I think it's just a testament as always to how Apple is slow,

Josh:
but they are figuring it out.

Josh:
And man, if they can get this foldable iPhone that turns into an iPad,

Josh:
runs models locally on your phone, it's gonna be pretty cool.

Josh:
So feeling bullish on Apple in general, I mean, what do you think this means

Josh:
for the valuation of the company, EJAS?

Josh:
Like Apple as a stock, is this, are we still being theoretical,

Josh:
hypothetical, or does this convert to actual revenue dollars?

Ejaaz:
I think it eventually converts to revenue dollars, but they're going to need

Ejaaz:
to deliver on, well, they're delivered on the hardware side.

Ejaaz:
They need to deliver on the software side. And that's typically where Apple

Ejaaz:
has really dominated the consumer market. Yes, they built an amazing iPhone,

Ejaaz:
But they also killed it with creating the best perfected app that you can use

Ejaaz:
on your phone and that entire ecosystem for developers and consumers on either side.

Ejaaz:
So they need to pull off the same thing for AI.

Ejaaz:
And the challenge that they're going to face is it's not the same as the internet that we know today.

Ejaaz:
It's going to be a new operating system. You and I have discussed multiple things on the show before.

Ejaaz:
We've discussed perplexity, releasing a new personal computer,

Ejaaz:
and then OpenAI, releasing an AI web browser. And all these random products

Ejaaz:
that you and I don't really use anymore.

Ejaaz:
It's used for very niche things. And what those attempts are getting at is trying

Ejaaz:
to rebuild an operating system around this new weird technology that kind of

Ejaaz:
feels like magic, but is also kind of dangerous.

Ejaaz:
And so Apple is the company that's currently being presented to solve that problem.

Ejaaz:
My bet is they're going to nail it at least within the next kind of two to three

Ejaaz:
years for two reasons. One, they have the largest distribution I was mentioning

Ejaaz:
earlier, 3 billion active devices.

Ejaaz:
So it's easy for them to kind of turn that on. Their biggest threat actually

Ejaaz:
might be Google, but Google doesn't really have the best experience of creating

Ejaaz:
consumer hardware, aka Google Glass.

Ejaaz:
There's actually a new version of that coming out in a few months time.

Ejaaz:
But yeah, I think Apple has the best shot of it.

Ejaaz:
Valuation wise, I actually think it's up from here. And they're currently valued

Ejaaz:
at what, like just under $4 billion or just over $4 trillion.

Ejaaz:
Dollars and i think they're going to see their kind of uh nvidia type rise now

Ejaaz:
um it wouldn't surprise me if they actually compete uh with nvidia for the top

Ejaaz:
spot once once consumer ai takes off.

Josh:
So in summary there is a lot of cool new hardware that just came down the pipe

Josh:
this week there is probably something for everyone and the noteworthy thing

Josh:
is this new kind of ai angle that they're taking they have the accessible devices

Josh:
starting at six hundred dollars they have the high-end devices like these new

Josh:
MacBook Pros that go up to $7,000, $8,000,

Josh:
but that are capable of running these really powerful, impressive local models

Josh:
on them. And is that going to

Josh:
be enough to actually start to detract away from the market shares of these

Josh:
other players like OpenAI and Anthropic. We will see.

Josh:
We will be monitoring the situation as always, but that is the news as it relates

Josh:
to Apple this week. It is a huge release in hardware.

Josh:
Are they accidentally sleepwalking into success or is this a tactical master plan?

Josh:
We don't know. Perhaps the new CEO, John Ternus, will tell us.

Josh:
But until then, thank you so much for watching. Like he just mentioned,

Josh:
he has a newsletter about this, about local edge inference that is releasing

Josh:
as you're watching this episode.

Josh:
It's out. So you can find that on our sub stack linked in the description below.

Josh:
Another way that you can help us is by sharing this episode with your friends.

Josh:
That always goes a long way. Share it with someone who you think will be interested.

Josh:
We do this four times a week, 25 minutes.

Josh:
If you listen to all of them, there's nothing that you will miss in the world

Josh:
of AI because our job is to keep you up to date. And thank you so much for staying up to date with us.

Josh:
And yeah, we will see you guys on the next episode.

Josh:
Hold on let me feed our system prompt we're you are going to record an absolute

Josh:
banger of an episode right now and you're going to make no mistakes think hard let

Ejaaz:
Me feed my biological llm you're going to absolutely smoke this episode and

Ejaaz:
it's going to go even more viral than the previous episode make zero mistakes.

Josh:
Let's go this is why we should charge premium all right i know