I Hope I'm Wrong About RAM Prices & AI Bubble

Let me be crystal clear.

I want to be excited here.

I want things to be changing and going back to the way they were.

Mark my words, though, and I hope I am wrong here.

Please, in a couple of months, come back to the video

and tell me how wrong I was

and how I should have been excited for everything happening.

We are not witnessing the AI bubble burst.

All we are seeing is resources being taken from one set of projects

over to another set of projects.

So, a few days back, OpenAI announced the end of Sora Video Generation.

This basically came out of nowhere publicly.

I'm sure behind closed doors they've been talking about it for a while,

but we'd not really heard anything about it.

There weren't any rumors.

It seems like this just pretty much happened overnight.

What's crazy about this, though,

is four months earlier, back in December,

we were talking about a deal between OpenAI and Disney

for $1 billion.

OpenAI gave no indication it was working to wind down Sora

in a blog post on Monday, so a few days prior,

titled, Creating with Sora Safely.

The company outlined ways it began working to make the app safe for teens

with stricter guardrails around harmful content

such as sexual material, terrorist propaganda,

and self-harm promotion.

This gives me the indication that

a lot of the people working at OpenAI

also weren't aware this was happening,

or maybe they just wanted to create some facade

just in case they weren't certain they were going to do so.

OpenAI told the BBC on Wednesday

that it has discontinued Sora so that it can focus on other developments

such as robotics that will help people solve real-world physical tasks.

A spokesperson for the Walt Disney Company said,

we respect OpenAI's decision to exit the video generation business

and to shift its priorities elsewhere,

which is the most corporate of corporate answer you could ever give.

Now, we don't have exact numbers,

but I think if the numbers floating around are anywhere near accurate,

it's pretty clear why they did this.

Sora wasn't making money.

In fact, people are estimating they were burning upwards of $10 million a day,

some estimates going as high as $15 million to...

do what exactly?

Generate videos of cats in ring doorbells doing funny things,

Ronald McDonald being arrested,

TikTok videos of religious events,

and other things which might be amusing, might be fun,

but...

can you justify burning millions of dollars every single day

just to generate memes?

Yes, there certainly was some use of it being used in advertisements

and various things like that,

but the majority of the usage...

was just...

slop.

It was brain rot.

It was...

it was just...

garbage.

But, I think much more importantly,

is the video generation had become very, very good,

and it was becoming a very easy target for lawsuits,

and there were already quite a few of those going on.

Then the growing concerns regarding deepfakes,

along with things like, you know, fake security footage,

and that footage...

if you've not seen it,

it's at a point where...

where a stupid judge,

a stupid jury,

could very easily be tricked by it.

If you're open AI,

maybe you don't have any morals about it.

However,

when you start counting up those lawsuits,

there comes a point where you have to say,

hmm,

are we actually losing money from the lawsuits now?

Does it make sense to continue doing this?

And where is this going to go?

Are the lawsuits going to slow down?

Or is this going to ramp up even more

as we deploy more of this,

as more people are using this,

as the tech gets better and better?

Following that,

a few days later,

we saw this.

Google has just announced TurboQuant,

a new AI algorithm

that reduces memory uses for large language models

by a factor of six,

and improves speed up to eight times via software.

I'll leave the white paper linked down below.

I do not have the time,

nor the technical expertise,

to explain how this actually functions,

and why this is better,

and why this is a good approach.

I'll leave this below.

Have a read of it yourself.

So,

here we have Google's announcement as well.

Introducing TurboQuant,

our new compression algorithm

that reduces LLM key value cache memory

by at least six times,

and delivers up to eight times speed up,

all with zero accuracy loss,

redefining AI efficiency.

Read the blog down below.

The non-technical explanation here

is more done,

less resources.

That's the part that matters.

Most likely,

as a direct result of that,

we have the stock graphs.

Now,

these are from a couple of days ago.

I'll show you the more up-to-date ones

in just a moment.

And for Micron,

it is not going amazingly for them,

at least in the short term.

And the same thing can be said

about SK Hynix as well.

Micron,

you might remember,

basically pulled back

on consumer RAM production.

So,

a lot of people are,

rightfully,

very happy

seeing the graph go like this.

However,

never forget that stock prices

do not directly reflect

the purchasing of a product,

or the price of the product.

They're this complex combination

of popularity,

purchases,

company drama,

the market as a whole,

your horoscope.

In other words,

vibes.

I did say the Micron graph

was a small excerpt

of what it actually looks like.

So,

this is the past day,

you know,

it's going down.

Past five days,

it's going down.

Past month,

it's gone down,

but notice this here.

Let's go the past year.

Yes,

it has come down.

And yes,

if you got excited about Micron,

like,

right here maybe,

you're probably not very happy.

But,

over the past year,

it's still massively skyrocketed.

This is not like,

a full correction of the market.

Micron just collapsing

to where it was before.

We're still very much

in the AI hype bubble.

What you're seeing is some,

sort of,

uncertainty.

Some questioning about,

how much of this RAM

is actually needed.

The exact same can be said

about Samsung Electronics.

Past five days,

gone down.

Past month,

again,

you have this spike here.

Past year,

it has still

massively skyrocketed.

And again,

SK Hynix,

past five days,