One month ago, the market was panicking

over the Iran war and broken [music]

supply chains. Today, it's verging on

extreme greed. But below all the

headlines, big institutions are quietly

building cash because they see a massive

opportunity ahead. My name [music] is

Alex and I spent eight years as an

electrical engineer and AI researcher at

MIT, which helped me find stocks like

Nvidia, Micron, Vertiv, and Coreweave

long before the rest of the market.

[music] In this video, I'm going to show

you the major market stories already

changing which stocks are about to win

big. And I'll show you a huge mistake

that I made along the way. Your time is

valuable, so let's get right into it.

The CNN Fear and Greed Index has been

stuck near extreme greed for the last

three weeks, the longest stretch so far

this year. And it's easy to understand

why everyone's so greedy, at least on

the surface. During earnings, the

biggest tech companies on Earth

announced over $700 billion in AI

infrastructure spending this year alone.

That's up 77% from last year, even with

their supply chains at a standstill. But

here's the signal that made me make this

video. Just last week at Berkshire

Hathaway's annual meeting, Warren

Buffett said the stock market is in a

mood for gambling, calling this

investing environment like going to a

church with a casino attached. Berkshire

Hathaway, under their new CEO Greg Abel,

is currently sitting on almost $400

billion in cash. That's a whopping 32%

of their entire portfolio. That's an

all-time record for Warren Buffett's

firm. That means they're holding more

cash than they did at the start of the

dot-com bubble, the global financial

crisis, and the pandemic. Warren Buffett

coined the phrase be fearful when others

are greedy and greedy when others are

fearful. And that's exactly what he's

doing right now. The question isn't why,

that part is obvious. War, broken supply

chains, and rising costs. The real

question is where that money goes when

he buys back in and when other massive

institutions follow his lead. Here's

where I think the biggest opportunities

in the market are right now.

Accountability is important to me, so

let's talk about my big mistake, because

I owe you a real apology here. Let's

talk about CPUs. Traditional AI data

centers run roughly one CPU for every

eight GPUs. The GPUs do the heavy math,

and the CPUs keep the traffic moving.

But a Jentick AI flips the script. When

a coding agent runs for 30 minutes

straight, it's making tons of separate

tool calls. It's spawning hundreds of

sub agents, and its memory usage can 10x

over the course of a session. None of

that orchestration, the tools, the sub

agents, and the context management runs

on GPUs. It all runs on CPUs. Jensen

Huang showed specific numbers at GTC

2026. 12,000 GPUs running at scale need

400,000 CPU cores running next to them.

That sounds like a 33 to 1 CPU to GPU

ratio, which is one reason why I think

stocks like AMD and Intel are running so

hot right now. But let me spend 30

seconds walking you through the real

math, and not the hype, and I'll show

you the mistake that made me miss a lot

of easy money. There are 72 Rubin GPUs

per rack. So 12,000 GPUs would need

about 167 racks. Each Vera rack has 256

CPUs, and each CPU has 88 cores. So one

rack of Vera CPUs has almost 23,000

cores. That means for every 167 GPU

racks, you actually only need 18 CPU

racks, or a ratio of 9 to 1 GPU to CPU

racks, not the other way around. In chip

terms, a data center wants 4600 Vera

CPUs for every 1200 Rubin GPUs, or one

CPU for every 2.6 GPUs. Now, here was my

mistake and why I owe you an apology. In

my opinion, investors should care about

the racks. After all, data centers are

built and priced in terms of racks. And

like I just showed you, a Gentex AI

needs nine times more GPU racks than CPU

racks. But, I didn't think about it in

terms of chips. And AI data centers only

need three times more GPUs than CPUs if

you count by chips instead of racks.

Said another way, there are roughly

three times more CPUs in AI data centers

than I realized. That's a big difference

and I really should have caught it

sooner. I didn't cover AMD enough and I

really should have. I didn't cover Intel

enough and I really should have. I

didn't listen to your feedback in the

comments and I really should have. I

apologize, full stop. So, let me put my

ego aside and cover AMD and Intel now.

AMD just reported earnings last week.

Revenues came in at $10.3 billion for

the quarter, which is up 38% year over

year. Data center revenue came in at

$5.8 billion,

which grew by a much higher 57%. Most of

that revenue comes from data center

CPUs, not GPUs. And their CPU sales are

growing a lot faster. But, like I just

said, those CPUs are much more important

in AI data centers than I realized. Meta

Platforms recently committed to 6

gigawatts of AMD's Instinct GPUs with 1

gigawatt for fully customized MI450s

built exclusively for Meta's workloads.

That's roughly the size of all the AI

compute on Earth combined outside of

Microsoft, Amazon, and Google. And Meta

just committed all of that to AMD. This

is a huge win for AMD, but it comes at a

huge cost that investors need to know

about. In order to make it happen, AMD

gave Meta a warrant, the right to buy

roughly 10% of the entire company at one

penny per share. Now, Meta only gets

those shares if AMD stock hits $600,

which would be around a trillion-dollar

valuation. But today, AMD already trades

over $450.

So, it's already 75% of the way there.

That means Meta gets almost a hundred

billion dollars in AMD stock essentially

for nothing. And AMD shareholders will

get diluted by 10%. But wait, they

actually get diluted by 20% because AMD

has the same deal with OpenAI. So, the

moment AMD stock touches $600 per share,

you take 20% off the top, and it's

actually only worth $480,

just $30 more than its current price.

That's the real cost of competing in

Nvidia's market. And there's something

else on the market that you need to know

about, and that's your private data.

There are hundreds of online data

brokers making big money by collecting

and selling your personal information.

That's why I've been using this video's

sponsor, DeleteMe, for over two years

now, and I can't recommend them enough.

DeleteMe is a hands-free subscription

service that will remove your personal

information from those online data

brokers. They give you a quarterly

privacy report showing everything

they've done. And they've reviewed over

55,000 listings for me so far. But what

really surprised me is that these data

brokers had way more than just my

private data. They had my wife's and my

entire family's, too. That's another

reason I really like DeleteMe. They have

a family plan, so we can all have more

control over our personal data. So, if

you care about your data and your

family's privacy, you can get 20% off

any consumer plan with my code Simple20

by going to join.deleteme.com/simple20

or with my link in the description. And

a big thank you to DeleteMe and to you

for supporting the channel. All right,

on their latest earnings call, Lisa Su

said that the data center CPU market

will nearly triple in size by 2030.

That's a compound annual growth rate of

35% or roughly three times faster than

the S&P 500. So, even with that 20%

dilution coming up, AMD's future is

looking pretty bright. But, while AMD

designs CPUs, Intel actually builds

them. For years, everyone said the same

thing about Intel, including me. Intel

needed a huge customer to prove that

they could still build chips and no big

company would take that risk. Until

then, Intel's foundry was a gamble, not

a business to invest in. But, everything

changed in April when Intel landed three

huge customers back-to-back-to-back.

First, Intel joined Terafab, a $25

billion chip factory being built in

Austin with Tesla, SpaceX, and xAI using

Intel's most advanced manufacturing

technology. For the first time, Intel

has a flagship customer lined up before

the factory was even finished. Then,

Intel signed a multi-year deal with

Google to build custom chips for their

internal cloud infrastructure. And just

a few days ago, Bloomberg reported that

Apple is in early talks with Intel and

Samsung about manufacturing their chips

in the US. That way, they can reduce the

risks of all their chips being made at

TSMC in Taiwan. If you didn't know,

Apple used Intel's chips in every Mac

from 2006 to 2020 when they switched

over to their own M1 chips made by TSMC

and Intel lost their biggest customer.

So, Apple coming back would be one of

the biggest comebacks in market history.

This is not a done deal, but it is the

first serious signal that Apple is

looking at chipmakers beyond TSMC for

the first time in over a decade. Intel

reported earnings a few weeks ago. Their

revenue came in at $13.6 billion,

which is up 7% year-over-year. And

earnings per share came in at 29 cents,

up from 13 cents last year. Those

numbers aren't too crazy, but the

market's reaction sure was. Intel stock

jumped 24% the next day, marking their

single best day since the dot-com era.

Two years ago, I said that Intel was a

value trap. Today, they're the only

American-owned and operated factory that

can build some of the world's most

advanced chips, and they might finally

have the customers to prove it. But, the

battleground for AI CPUs just got a lot

bigger. For the last 35 years, ARM was

the arms dealer that never picked a

side. They sold blueprints to Nvidia and

AMD, Apple and Qualcomm, and they

collected royalties while everyone else

fought the actual chip war. That is,

until now. A few weeks ago, ARM launched

the AGI CPU, the first chip they've ever

designed and sold themselves in the

history of the company. And right off

the bat, the specs are pretty serious.

ARM's AGI CPU has up to 136 cores per

chip and 60 chips per rack. So, let's do

the same math we just did a few minutes

ago and compare it to Nvidia. Remember,

according to Jensen, it takes around

400,000 CPU cores to support 12,000

Rubin GPUs. 136 cores per ARM chip * 60

chips per rack is just under 8,200 cores

per rack. So, it would take around 49

racks to support those GPUs compared to

just 18 racks of Nvidia Vera CPUs. That

sounds way worse, but let me fix the

mistake I just showed you I kept making

instead of falling into the same trap.

If we look at the actual chip counts

instead of just the racks, it takes

4,600 Vera CPUs to support those 12,000

Rubin GPUs. That's one CPU for every 2.6

GPUs, just like I showed you before.

But, you only need 3,000 ARM AGI CPUs to

support that same amount. That's one CPU

for every four GPUs, or around 54%

better performance than Nvidia's Vera.

Said another way, arm's new CPU is

actually much more powerful than

Nvidia's. So much so that you need

almost 40% fewer to run the same data

centers. And that's just versus Nvidia.

It has roughly double the performance

per watt versus Intel and AMD. And arm

says it can save around $10 billion in

construction costs per gigawatt of data

center compute. So let's bring

everything full circle. Meta just

committed to building 6 gigawatts in

data center infrastructure using AMD's

chips. If they used arm's AGI CPUs

instead, they would have saved $60

billion on this project alone, which is

pretty close to what AMD paid Meta to

win that deal in the first place. And

I'm not the only one who caught that

math. Meta did, too. That's why they're

arm's first and flagship customer for

the AGI CPU. And early demand for this

chip is through the roof. As soon as arm

started taking orders, their demand

doubled in the first six weeks. And on

their earnings call just a few days ago,

arm's CFO said they expect to sell over

a billion dollars worth of these CPUs in

the first year alone, and hit $15

billion in annual chip revenue by 2031.

The entire company makes less than $5

billion a year today. So arm is

expecting this chip to quadruple their

annual revenue over the next five years.

So arm didn't just enter the CPU

battleground. They dropped a tactical

nuke on it. And the AGI CPU is only one

part of their story. Arm's royalties

from data center chip designs more than

doubled year over year, beating every

estimate. And these royalties have

insane 95% gross margins, much higher

than even the best software companies,

let alone hardware firms. The reason the

stock dropped 10% after their earnings

goes back to what I said at the start of

this video. Supply chain issues are

stopping them from growing even faster.

Arm's problem isn't that nobody wants

their chip, it's that they can't build

those chips fast enough and the smart

money knows it. But if you think demand

for AI infrastructure is insane right

now, there's one more bombshell I need

to walk you through. Now, to be clear,

what I'm about to show you isn't

verified, so it could be nothing or it

could change everything. And if you

found this video valuable, consider

hitting the like button and subscribing

to the channel. It really does help and

it tells me to make more content like

this. All right, let's talk about what

could be the single biggest breakthrough

in AI efficiency since the original

Transformer paper. Last week, a Miami

startup called Subquadratic announced a

new model called SubQ 1M preview. The

research version of this model has a 12

million token context window, which is

up to 12 times bigger than most frontier

models today. The version they plan to

actually ship matches everyone else at 1

million tokens, but it claims to be over

300 times cheaper to run. Said another

way, if you spent $2,500 doing work on

Claude, you could do that same work on

SubQ's model for the price of a cup of

coffee. They raised $29 million in seed

funding and launched at a $500 million

valuation. But here's what makes this

announcement pretty hard to judge. It

didn't come with a peer-reviewed paper,

it didn't come with a public model to

test, and it didn't come with any

benchmarks that anyone outside the

company could reliably reproduce. So,

I'm watching what happens next. If

Anthropic, OpenAI, or Google publish

their own work on subquadratic attention

in response, that means these claims are

being taken seriously. But if there's

crickets, this is probably just smoke

and mirrors. But either way, investors

need to understand what happens next if

this turns out to be real or there's

another breakthrough just like it. When

Dwave came out, AI got dramatically

cheaper overnight, but instead of demand

falling, it exploded because every

dollar of compute went a lot further.

When the cost of something drops, you

get much more bang for your buck. So,

overall demand goes way up. Cheaper

smartphones don't mean people use less

data. It means more people, more apps,

and more content being consumed every

single day. And more data centers had to

be built to handle it. So, if sub Q

turns out to be real, every piece of the

AI revolution becomes even more

valuable, including all the CPUs I made

the mistake of ignoring before this

video. A mistake I won't make again. At

the start of this video, I pointed out

that CNN's Fear and Greed Index has been

close to extreme greed for the last 3

weeks, the longest stretch so far this

year. But, Warren Buffett is sitting on

record levels of cash. Warren Buffett

coined the phrase "Be fearful when

others are greedy and greedy when others

are fearful." And that's exactly what

he's doing right now. The question isn't

why. We already know that. War, supply

chains, and rising costs. The question

is where that money goes when big

institutions buy back in. And now, you

know that, too. And if you want to see

what else I'm investing in, 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.