Something big is happening at Google and

Amazon. They just launched chips to

challenge Nvidia's data center

dominance, which could spell big trouble

for the world's most valuable company

and change the course of the entire AI

revolution. Your time is valuable, so

let's get right into it. About a week

ago, news broke that Meta Platforms was

in talks to spend billions of dollars on

Google's custom TPU chips. But instead

of rushing out to make a video, I took

some time to understand what these chips

actually do and what this means for the

AI market. Because in my opinion, the

single most important question for AI

investors is how long Nvidia can

dominate data centers with their GPUs

and CUDA ecosystem since so much of the

AI market is built on top of them today.

And I'm glad I waited because Amazon

also announced their new tranium 3 chips

a couple days ago. So, I'll break those

down for you, too. I'm also not here to

hold you hostage. So, here's exactly

what I'll cover in this video. I'll

explain what Google and Amazon's chips

actually do, how they compete with

Nvidia's hardware ecosystem, how these

chips actually could change the course

of the entire AI revolution, but not in

the way most investors think, and of

course, which AI stocks I'm buying as a

result. There's a ton to talk about. So,

let's start with the story that's on

everybody's mind. About a week ago,

Google announced that they would sell

their custom tensor processing units or

TPUs to other data centers. According to

Morgan Stanley, Google has a roadmap to

ship a million TPUs to external

customers by 2027, which would increase

their cloud revenue by over 10% or close

to $13 billion. Google's long-term

internal goal is to capture around 10%

of Nvidia's data center revenues over

time, which works out to tens of

billions of dollars every year. They

plan to do that by chipping away at some

of Nvidia's biggest customers and their

most widely supported workloads. This is

a huge change from Google's previous

chip strategy, and it's actually much

bigger than just competing with Nvidia.

Let me show you why. While GPUs are

generalpurpose accelerators that can run

almost everything, TPUs or tensor

processing units are a different kind of

chip called AS6, applicationspecific

integrated circuits. Google's TPUs are

specifically built for tensor operations

like matrix multiplication and the

related maths that dominate deep

learning today. That makes Google's TPUs

especially good at three kinds of

workloads. First, they're great at high

volume inference at massive scales.

Google serves billions, if not trillions

of requests across very specific

services like search and advertising,

maps and shopping, YouTube and Gemini.

So, their biggest hardware bottleneck

isn't flexibility, it's efficiency.

Google's TPUs can outperform GPUs by

anywhere from 50 to 100% per dollar or

per watt, but only for this specific set

of applications. Their performance also

scales extremely well when thousands of

TPUs are connected together for parallel

computing thanks to each chip having

integrated networking, fast

interconnects, and being tightly coupled

with memory. That also makes them great

at large training jobs for those same AI

models. And the third kind of workload

that Google's TPUs are great at are

specialized recommendation and ranking

systems. Since so many of their services

involve ranking websites, videos,

products, businesses, and advertisements

based on searches, demographics,

browsing, and purchase history, and so

on. Google's TPUs have custom hardware

to accelerate data requests from huge

lookup tables and do the highly

specialized math involved in ranking the

results. There are a few key reasons

that Meta Platforms would want to buy

these TPUs. First, both companies have

very similar workloads. Google has

YouTube, Meta has Instagram, Google has

Gemini, Meta has Llama, and so on. And

it's cheaper to buy Google's TPU based

AI factories than to build their own

full stack solutions from scratch. Not

just the chips, but the racks, the

liquid cooling, optical interconnects,

workload schedulers, and software that

all need to be designed together. Also,

Meta's training and inference

accelerators, also known as the MTIA

chips, only support a limited amount of

workloads, mainly focused on inference.

While Google already has pods of 10,000

TPUs training frontier scale models for

search, for video, and for large

language models, making them a great way

for Meta to catch up on the AI hardware

race and diversify their hardware

portfolio beyond Nvidia's GPUs. But

there are a few important points that

investors should understand about this

potential deal. Meta is spending around

$70 billion on AI infrastructure this

year alone and their capex budget for

2026 is projected to be close to a

hundred billion. That's a massive amount

of internal demand that their MTIA chips

can't possibly fill. But other

hyperscalers don't have this same

problem. While Amazon and Microsoft also

have massive capex budgets, their custom

AI chips and internal hardware systems

are much more mature than Metas. So,

they're much less likely to buy Google's

TPUs instead of just investing in their

own already proven technologies. In

fact, Amazon Web Services just launched

their new Tranium 3 chip. Another ASIC

that's focused on extreme power

efficiency and cost savings for a few

specific AI workloads that they run at

extremely high volumes like training and

inference for large language models with

huge parameter accounts and context

windows as well as the multimodal and

mixture of experts models behind

powerful AI agents like Claude. This

chip has 50% more memory capacity, 70%

more bandwidth, twice the compute

performance, and is 40% more energy

efficient than Amazon's previous

generation. So, at first glance, 2026

could be a very tough year for Nvidia.

In fact, 2026 is right around the

corner. And while everyone is making

resolutions they won't keep, you can

learn AI before the new year even

begins. That's where Outskill comes in,

the sponsor of this video. Outskill is

hosting a 2-day live AI workshop to take

you from beginner to AI power user this

Saturday and Sunday. Over 10 million

people around the world have already

attended and the slots for this one are

filling up faster than ever. But for the

holidays, they're giving the first 1,000

people who sign up with my link a free

seat. Whether you work in tech or

finance, marketing or retail, you'll

learn to use some of the most powerful

AI tools, best practices for prompting

and using AI agents, and even how to

automate workflows without any coding.

You'll also get free access to Outskills

online community and learning dashboard

for people turning artificial

intelligence into real revenues. This is

a great way to gain a competitive

advantage, get a serious head start for

2026, and understand the science behind

the stocks. So, make sure to register

for your free seat before they run out

with my link below today. All right. So,

both Google and Amazon have high

performing custom chips that go after

the same part of the AI market. Training

and inference for huge high throughput

AI models and recommener systems that

need to handle billions of consumer

requests. And now that we understand

what these chips do, let's talk about

how they actually compete with Nvidia's

hardware ecosystem. The truth is they

mostly don't. Outside of those very

specific workloads, Nvidia's GPUs power

many kinds of AI across a wide variety

of industries. Not just token generation

for large language models, but image and

video generation, physics modeling and

simulation, professional visualization

and product design, protein folding and

drug discovery, robotic motion and

self-driving cars. The list goes on and

on. And if you've been watching this

channel for a while, you know that

Nvidia's hardware ecosystem is much

bigger than just GPUs. In fact, it

includes something called NVLink Fusion.

NVLink Fusion is a special chiplet that

can be added to other CPUs or other

accelerators so they can be installed in

Blackwell's compute trays or so that

Blackwell's GPUs and networking

solutions can be used in data centers

that are already invested in other chips

like ARMbased CPUs or more application

specific accelerators. So, while

Google's TPUs might be more powerful for

specific workloads, they're also more

closed, forcing data centers to rely on

Google's hardware and software stack as

is. And Google's stack is nowhere near

as versatile or as widely adopted as

CUDA. That's why Google's long-term goal

is to capture only around 10% of

Nvidia's GPU market with their TPUs,

like I mentioned earlier. And Amazon's

total addressable market is even smaller

since they're keeping their Tranium 3

chips inhouse, which means companies

will have to run their workloads on AWS

if they want to use these chips. On the

flip side, Nvidia has millions of GPUs

in almost every AI data center on Earth

from AWS and Google Cloud to Microsoft

Azure and Meta Platforms' AI

superclusters. And now that we

understand how Google's and Amazon's

chips compete with Nvidia's hardware

ecosystem, let's talk about what this

all means for the AI revolution and

which stocks are worth buying as a

result. And if you feel I've earned it,

consider hitting the like button and

subscribing to the channel. That really

helps me out and it lets me know to make

more content like this. Thanks. And with

that out of the way, let's talk about

how these chips impact the overall AI

market. First, the AI market is

projected to almost 19x in size over the

next 9 years, which would be a compound

annual growth rate of over 38% through

2034. That's almost three times faster

than the average growth rate of the S&P

500. So, even if Google does take 10% of

Nvidia's market share over time, that

market is growing much faster than

either of them can fill the demand

alone. But that growth is also spread

across very different areas of AI like

natural language processing, computer

vision, autonomy, and robotics. Nvidia's

GPUs are flexible enough to support all

of these different segments. While

Google's TPUs and Amazon's Tranium chips

focus on machine learning and natural

language processing. So, while Google

and Amazon might challenge Nvidia's

pricing power and their margins with AI

labs and AI data centers focused only on

language models like OpenAI and

Anthropic, they can't really compete

once robots, sensors, physical motion,

or digital simulation enter the

equation. And don't forget, the biggest

AI labs and data centers won't ever go

allin on one ecosystem anyway since they

don't want to be dependent on a single

vendor and their clients want access to

the best chips at the best prices, which

change depending on the workload. That's

why every hyperscaler has their own AI

chips in the first place. Now, let me

say the quiet part out loud. The biggest

loser in this situation is AMD since

their entire data center strategy is

being the cheaper, better bang for the

buck alternative to Nvidia, especially

for large language model inference,

which is exactly what Google's and

Amazon's new chips are designed to do.

Google's TPUs will compete directly with

AMD for costoptimized inference

performance and for some specific

training workloads in Google Cloud, in

Meta's data centers, and with Anthropic.

and Amazon's Tranium 3 chips will lower

the need for AMD's GPUs inside AWS. So,

as more companies build more application

specific chips, they will reduce

Nvidia's pricing power and margins, but

they could remove the need for AMD's

chips altogether. All right, but who are

the biggest winners in this situation?

The Taiwan semiconductor manufacturing

company, ticker symbol TSM, is the only

company on Earth capable of making

Nvidia's GPUs, Google's TPUs, and

Amazon's Tranium chips. They also

manufacture Microsoft's custom Maya

accelerators and Meta's training and

inference chips. So, as their customers

start designing more specialized chips

for different kinds of workloads, there

will be even more demand for TSMC's most

advanced and most profitable chip

production nodes. On top of that, more

specialized chips require more advanced

packaging techniques to get the

processors, the memory, and the

networking components close enough

together and connected at ultra high

bandwidths. Not only is TSMC the market

leader by far when it comes to advanced

packaging, it's also the main driver for

their margins. So, long story short, the

more demand there is for different kinds

of AI chips, the more TSMC can charge

for their limited supply, which makes

TSM a great stock regardless of who wins

between Nvidia or Google. Amazon or AMD.

But we can't talk about custom chips

without talking about Broadcom. Ticker

symbol AVGO. Broadcom helped design

multiple generations of Google's TPUs,

Meta's Training and Inference

Accelerators, and Bite Dance's custom

chips that help power Tik Tok. And just

a few weeks ago, Broadcom announced a

massive partnership with OpenAI to

design their XPUs, which are custom

processors optimized to power Chat GPT,

GPT5, and OpenAI's future models. But

regardless of which accelerator wins the

AI era, those chips need to be connected

with ultra-igh speeded networks, which

is another area where Broadcom competes

directly with Nvidia. In fact, Broadcom

has a 90% market share in Ethernet

switching chips for data centers, which

is just as huge as Nvidia's share of the

data center GPU market. Around 30% of AI

workloads run on Ethernet today. And

that number is actually growing since

the overwhelming majority of the world's

data centers already run on Ethernet.

That's why Nvidia also offers

Ethernet-based networking products

instead of forcing data centers to

switch to Infiniband which they own. So

by holding both Broadcom and Nvidia

stock, investors are holding the two

companies selling networking solutions

to almost every AI data center and

supercomputer in the world. But the

whole reason that Google and Amazon even

bother designing their own custom chips

is for power efficiency. That's because

electricity accounts for around onethird

of a data cent's ongoing operating

expenses and cooling accounts for about

40% of that. That means both power and

cooling have a huge impact on the profit

margins for AI. Which is why the third

stock on my list is Vertive Holdings,

ticker symbol VRT. Verive Holdings makes

power and cooling systems for data

centers. For example, they make liquid

cooling systems specifically for

highdensity servers and massive GPU

clusters used for AI training and

inference. Verdives liquid cooling

systems are modular in design and they

can be scaled up to cool 600 kW worth of

server racks per unit. That means that

one of these systems can cool five

kowatt Blackwell racks without a data

center needing to overhaul its

pre-existing infrastructure. As a

result, Vertive supplies cooling

solutions to all three cloud

infrastructure providers, AWS, Google

Cloud, and Microsoft Azure. Verdive also

supplies them with core power systems

like their Liber XL, which is a

high-capacity, uninterruptible power

supply designed specifically for hypers

scale and cloud facilities, and it

supplies huge amounts of energy at very

high efficiencies. So hopefully this

video helped you understand what Google

and Amazon's custom chips do and their

overall place in the AI revolution,

competing with Nvidia on specific kinds

of workloads, but threatening AMD's

position as a cost-effective alternative

in the process. And of course, three

stocks set to win big regardless of

which AI chip actually comes out on top,

making them all a great way to get rich

without getting lucky. And if you want

to see even more stocks that 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 Tickerol U. My name is Alex,

reminding you that the best investment

you can make is in you.