If you put $10,000 into Nvidia when Chat

GPT launched less than 4 years ago,

you'd have over $120,000

today. If you invested that money into

Micron just last year, you'd already

have over 75 grand. Well, there's

another kind of AI chip that's quietly

changing which stocks are about to win

big and making their investors very rich

along the way. My name is Alex and I

spent 8 years as an electrical engineer

and AI researcher at MIT. And I never

thought that this would be the next big

battleground for AI. Let me show you

what's going on and how I'm investing in

it. Your time is valuable, so let's get

right into it. For the last 4 years,

data centers were priced as a GPU

business. Accelerators from companies

like Nvidia, AMD, and Broadcom are the

product and every piece of AI

infrastructure was built around them.

Not just the power, the cooling, and the

server racks, but also the networking

chips, the memory, and even the CPU

alongside them. But that's quickly

changing and to

we need to understand the science behind

the stocks. GPUs break math problems

down into thousands of tiny pieces,

solve every piece at once, put the

pieces back together, and return the

final answer. That's why they're so

fast. But the trick is that not every

calculation can be broken into tiny

pieces in the first place. That's why

many of Nvidia's CUDA libraries focus on

transforming real-world problems into

problems that a GPU can solve. That's

what they mean by accelerating different

industries. A CPU works the exact

opposite way. It can handle any kind of

problem, but it handles one problem at a

time. So, GPUs do predictable work in

parallel while CPUs do unpredictable

work in sequence. When you ask Chat GPT

or Claude or Gemini a question, it

basically goes straight to a GPU. Your

sentence gets chopped into tiny pieces

called tokens and every piece runs

through the same neural network.

Billions of numbers, each multiplied in

the same pattern, layer by layer, the

perfect shape for a GPU. So, the CPU

hands the work over and then stays out

of the way. But, AI agents flip this

script entirely. An agent reads your

request, comes up with a step-by-step

plan, makes a bunch of separate tool

calls, writes code, and maybe even

spawns sub-agents to handle different

steps along the way. The math still runs

on GPUs, but everything else runs on the

CPU. Choosing what to do next, fetching

files, moving data, and so on. Even when

I worked on these kinds of systems back

at MIT 10 years ago, which were way

smaller and simpler at the time, the GPU

was waiting while the CPU had to think.

Agents push that to the limit, which is

why the CPU is becoming the next big

battleground for AI. And the battle is

between the biggest companies with the

most advanced technology in the world.

Let's start with Nvidia. Nvidia has

their Grace CPUs and superchips. The

Grace CPU has 72 cores, and the

superchip connects two of them together,

so they act like one giant CPU with 144

ARM Neoverse V2 cores. But, here's the

big problem. Neoverse is the data center

CPU core design that's sold by ARM.

Nvidia only licenses it. So, Nvidia

might be the king of AI CPUs, but the

castle belongs to ARM. Nvidia sold

nearly 2.5 million Grace CPUs to date,

and every one of them pays a royalty

back to ARM. Nvidia's answer is a chip

called Vera, which has 88 custom Olympus

cores that are designed in-house, making

it their first fully custom CPU. Nvidia

kept exactly one thing from ARM, the

instruction set. That's the language the

processor understands, and the list of

commands that software can give it. So,

code written for Grace can run on Vera

without needing to be rewritten. That

means Nvidia can keep every line of code

their customers already wrote and stop

paying arm for the cores that run it.

Nvidia already delivered the first Vera

CPUs last quarter to Anthropic, OpenAI,

SpaceX, and Oracle. On their latest

earnings call, Nvidia reported over $75

billion

in data center revenues for the quarter,

roughly $60 billion from compute and $15

billion from networking. When investors

think about AI compute, they think about

GPUs. But Nvidia's CFO, Colette Kress,

said that they expect around $20 billion

in CPU revenues this year alone and that

AI CPUs are a brand new $200 billion

total addressable market for Nvidia. But

Nvidia doesn't dominate this market the

same way they dominate the GPU market.

In fact, they have some serious

competition from arm, from Qualcomm, and

from AMD. All three are massive chip

companies and all three are making some

very big moves. By the way, if you've

ever wondered whether a stock is

actually worth its price, you are not

alone. Finding a company's fair value is

one of the hardest parts of investing.

That's why I use ProPicks by

Investing.com, the sponsor of this

video. Each month, their Tech Titans

strategy narrows the entire US tech

sector of the market down to just 15

high-quality stocks. The algorithm finds

companies with the best mix of

fundamentals, momentum, and guidance.

When the data changes, the strategy

updates, staying only in the highest

conviction stocks. It also gives you the

reasons behind every move, not just what

changed, but why. You can click into any

stock and see a full breakdown of the

data to kickstart your own research. But

the proof is in the pudding. Since going

live, Tech Titans has more than doubled

the S&P 500, returning over 180% in less

than 3 years. Today, it's holding great

stocks like AMD, Applied Materials,

Qualcomm, and ARM, all of which I cover

on this channel. And right now,

investing.com is running a summer sale,

which is their best price all year. And

you can use my link below to get another

15%. Talk about a no-brainer. All right,

let's start with ARM, since they own the

architecture that NVIDIA's CPUs

currently rely on. On July 29th, during

their earnings call, ARM told

shareholders that the initial version of

their AGI CPU was already delivered to

multiple customers, marking the end of a

35-year era. Ever since they were

founded in 1990, ARM was the arms dealer

that never picked a side. They sold

blueprints to Apple and Qualcomm, NVIDIA

and Amazon, and they collected up to 5%

royalties on every chip, while everyone

else fought the war. That is, until now.

ARM recently launched the AGI CPU, the

first chip they've ever designed and

sold themselves, which means they make

100% of the revenue instead of just

five. And right off the bat, this chip

has some serious specs, up to 136 cores

per chip, running at 300 W, and built on

TSMC's 3 nm process, with Meta as the

lead partner and co-developer. ARM

claims that their AGI CPU has more than

double the performance per rack versus

x86 chips, and they have over $2 billion

of demand over the next 2 years. But,

they've only secured enough

manufacturing capacity to meet around

half of that. That capacity is being

constrained by wafers, substrates,

memory, and testing, all things that ARM

never had to worry about when they were

only selling blueprints. And that shift

is already showing up in their numbers.

Revenues came in at $1.3 billion for the

quarter, which is up 22% year-over-year,

with revenue from royalties also up by

22% but data center royalties more than

doubled growing at least four times

faster than the rest of the business.

This was also the quarter where arm

stopped reporting remaining performance

obligations. That's the revenue they

have under contract but haven't

collected yet which usually gives

investors visibility into their next few

quarters so they can track and predict

arms growth. In my opinion this was the

single most useful forward-looking

metric that arm published and they

stopped the same quarter that they

started shipping hardware. The last

number that we got was 2.07 billion

dollars at the end of March which is

actually down 7% from the year before.

But now we have to wait until their

annual report next spring for the most

updated number. Until then I think gross

margin is the number to watch. Arm went

from collecting high margin royalties

with very clear visibility to also

having to worry about wafer costs,

production capacity risks, and hardware

like margins. Investors should see real

revenues from the AGI CPU in fiscal year

2028. So that's when we'll get a much

better sense for their overall margins.

But I think the market is already

pricing in their chip sales to

perfection. Arm's current price to sales

ratio is close to 50. Not earnings,

sales. That's almost 10 times higher

than the semiconductor industry average

and their forward price to sales ratio

is almost 39 while every other stock in

this video sits between 4 and 14.

Qualcomm is one of Arm's biggest

customers and their forward price to

sales ratio is under four. So let's talk

about them next. If you own an Android

phone there's a good chance that

Qualcomm built the processor inside it.

Their Snapdragon chips are in most

flagship Android devices. Their modems

were inside iPhones for years and they

collect some kind of royalty on almost

every smartphone on the planet even the

ones running on someone else's silicon.

They even make some chips for cars, but

Qualcomm builds their Orion cores under

an architectural license from ARM.

Qualcomm also reported earnings on July

29th. Revenues came in at $9.9 billion

with $2.21 of earnings per share. Their

automotive business hit a record $1.6

billion,

which is up 61% year-over-year. That's

their 23rd straight quarter of

double-digit automotive growth. Internet

of Things came in at $1.8 billion, which

was up 9%, but handsets came in at $5.1

billion, which was down 20%

year-over-year. That decline is largely

due to them having lower share of modems

in the coming iPhone launch as Apple

keeps transitioning to their own custom

chips. Qualcomm's management said that

this quarter marks the bottom for

handsets and announced double-digit

price increases on their chips starting

September 1st. That also means the

non-phone business is now carrying the

company. Qualcomm guided year-over-year

growth in non-handset revenue to jump

from 24% this fiscal year to over 60%

next year, and they specifically called

out data centers as part of that growth.

At their investor day back in June,

Qualcomm set a target of more than $15

billion in data center revenue by 2029.

This year, it's around $300 million.

So, that's roughly 50x growth in just 3

years, and it would account for more

than a third of Qualcomm's entire

trailing 12-month revenue of $44

billion. Said another way, Qualcomm

expects their new data center segment to

account for roughly a third of

everything it took them 40 years to

build so far. Even though their data

center revenue is small today, they've

already lined up some massive customers.

Meta signed a multi-generation agreement

for Qualcomm's CPUs for their

next-generation server fleet, and Satya

Nadella confirmed that Azure will deploy

Qualcomm's accelerators starting next

year. Qualcomm's Dragonfly C1000 at data

center CPU also has some serious specs,

over 250 Orion cores across multiple

connected dies, kind of like Nvidia's

Blackwell GPUs. But, here's the thing.

Qualcomm has never shipped a data center

CPU. Their custom silicon revenue starts

later this year. The Azure accelerators

land in 2027, and the C1000 CPU itself

doesn't start production until the

second half of 2028. So, the money is

booked, the customers are signed, but

the actual chip is still 2 years out.

Like I said earlier, Qualcomm's forward

price to sales ratio is under four, 10

times cheaper than ARM. So, while ARM's

AGI CPU is already priced in, Qualcomm's

C1000 CPU seems to be priced at zero,

making it a good opportunity to get in

early, at least for long-term investors

that are willing to take a risk on a

brand new chip from a company that just

pivoted to data centers. But, if you

want to bet on the current king of CPUs

instead, let's talk about AMD next. And

if you feel I've earned it, consider

hitting the like button and subscribing

to the channel. It really does help, and

it lets me know to make more content

like this. Thanks. Now, let's talk about

AMD's AI chips. On July 20th, Microsoft

announced that they'll deploy AMD's

Helios systems on Azure, pairing the

Instinct MI455X

GPUs with their sixth generation Epic

CPUs, code named Venice. Let me break

that down for you. Helios is a

rack-scale system, which means that AMD

is selling an entire rack, accelerators,

CPUs, networking, and software, all

wired up and optimized to work together,

delivered as a single unit. Microsoft is

coming out with three new kinds of

virtual machines based on AMD's

hardware. The NDMI455X

V7 for AI inference on Helios itself,

the Azure HDV2 for a gentic AI and data

pipelines, and the Azure HX V2

specifically for semiconductor design.

Here's why this is a big deal. Cloud

providers only create and offer virtual

machines at scale for configurations

that they plan to run and rent out for

years. So, Microsoft is basically

committing to AMD's racks much more

aggressively than any normal purchase or

could. They're putting AMD's technology

on the menu as a permanent product line.

Helios is already in production and

leading AI companies are going to deploy

these systems at gigawatt scales. Just

to be clear, 1 gigawatt is roughly the

power output of a large nuclear reactor.

Open AI signed a deal for 6 gigawatts

worth of AMD GPUs. Meta signed a deal

for 6 more gigawatts of Helios racks

running Venice. Anthropic signed up for

another 2 gigawatts and AMD committed up

to $5 billion worth of equity to

Anthropic that same week. And here's

where we come full circle. AMD's Helios

racks have 72 distinct MI455X GPUs and

18 epic Venice CPUs. That's one CPU for

every four GPUs. Almost every Wall

Street analyst models GPU sales, but

almost none of them account for the 18

powerful and expensive CPUs that are

shipping with every rack. CPUs that are

becoming more and more necessary to

power AI agents and CPUs that have zero

competition because Helios comes as one

integrated package. I think the entire

way we measure and predict market share

is about to change. The old way is

counting units. How many chips or

dollars every company sold versus their

competitors. But, the new way is

counting tokens and watts. Each

company's share of AI power deployed and

the amount of tokens their systems are

generating per watt. OpenAI, Meta, and

Anthropic are signing up for 14

gigawatts under AMD. That's a pretty big

deal if it all actually gets deployed.

But, there are three important details

to the AMD story that investors really

should remember. First, AMD is paying a

serious sum for those gigawatts. AMD

gave OpenAI warrants for up to 160

million shares, and Meta Platforms 160

million more, all priced at 1 penny

each. That means they're giving away up

to 20% of the entire company to secure

these deals, and they invested another

$5 billion into Anthropic on top of

that. Second, these deals are not

guaranteed. There are a lot of

performance milestones along the way,

and OpenAI and Meta still have to

actually deploy these chips in racks.

So, all of that is still in front of

them, even though it's all already

priced into AMD's stock. And third, AMD

just reported earnings right as I'm

recording this video. The revenue came

in at $11.5 billion,

which was up 50% year-over-year. Data

center revenues came in at $6.7 billion,

up 107%

year-over-year, and now accounting for

58% of AMD's total revenues. They even

guided next quarter's revenue to roughly

$13 billion. That sounds great until you

compare it to Nvidia. I know I'm beating

a dead horse here, but it's important to

know that AMD is five times more

expensive than Nvidia by PE ratio, and

three times more expensive by forward PE

ratio. And AMD's multiples have only

been going up with each earnings call,

while Nvidia's have been coming down.

I'm not saying you shouldn't buy AMD

stock, but I am saying you should

understand what you're buying if you do.

Hopefully, this video helped you

understand that AI CPUs are no longer

stand-alone chips. They're a core

component of the infrastructure for

agentic AI and a big battleground for

every major semiconductor company. And

AMD's CEO Lisa Su just confirmed on

their earnings call that demand for

accelerators and CPUs keeps growing

faster than anyone expected. Funny

enough, the two companies sitting at the

center of it all are actually Meta

Platforms and TSMC, neither of which

sell any CPUs themselves. Meta is Arm's

lead partner for the AGI CPU. A flagship

customer for Qualcomm's Dragonfly C1000,

they bought 6 gigawatts worth of Helios

racks from AMD, and they're one of

Nvidia's largest AI infrastructure

customers. So, if you hear that AI

spending is slowing down, Meta is the

company to check first. And all four

companies rely on TSMC to make these

chips in the first place. AMD's Venice,

Nvidia's Grace and Vera, Arm's AGI CPU,

and Qualcomm's Dragonfly C1000 are all

made on TSMC's most advanced nodes. So,

if they have any delays, supply

constraints, or somehow fall behind,

every company we just covered will feel

it, too. But as for me, I'm betting that

Meta spending keeps growing, that TSMC

keeps producing, and that AI agents are

here to stay, making data center CPU

stocks a great way to get rich without

getting lucky. And if you want to see

even more stocks I'm buying to get rich

without getting lucky, check out this

video next. Either way, thanks for

watching, and until next time, this is

ticker symbol U. My name is Alex,

reminding you that the best investment

you can make

is in you.