An insightful financial analysis examining why memory stocks pulled back despite strong fundamental catalysts in the semiconductor industry (1:42). The video breaks down TSMC's record-breaking August revenue growth, the escalating high-bandwidth memory (HBM) supply shortage impacting global AI compute prices, and SK hynix's newly unveiled Full-Stack AI Memory strategy designed for agentic AI architectures.#Micron#SKHynix#SanDisk#TSMC#HBM#DRAM#NAND#AIStocks#Semiconductors#Investing
AI Generated·May contain errors
Video Summary
AI Generated
Despite short-term market volatility and macro jitters surrounding the CPI report, the structural bull case for AI hardware remains strong. Record revenues from TSMC and severe HBM shortages in China highlight a critical demand surge. The industry is shifting toward a \"full stack\" memory architecture to support the evolution from generative to agentic AI, increasing the value of all memory formats.
Coming to you on location again. I'm traveling right now. I actually just got into Florida.
I'll be here in Palm Beach all through the weekend heading back to Texas on Sunday,
but the next couple of videos will be coming to you from sunny Florida here in Palm Beach,
a spectacular part of the country. I also want to give a public service announcement before we get
into tonight's video, because I became aware through a member of our community that there
is someone that is falsely reaching out, falsely misrepresenting themselves as me.
They're using my name, they're using the logo, and they're sending emails to members of this
community. Guys, know that this is a complete scam. I will not reach out proactively and ask
you to invest in certain things, to shift capital, to join elite clubs or private clubs. It's a
complete scam. Please ignore these messages. I also will make sure I drop my email below in the
description. If anyone has reached out to you and you want to confirm it's fake or it's true,
feel free to send me an email. I'm happy to respond and let you know if anything is accurate,
but I have not reached out. If you've received anything from a Kevin.Garrity at a Gmail account
using my logo, most likely it is a scam and I would highly recommend that you ignore it.
So thank you for those in the community that brought this to my attention. Know that I'm
hard to remove the scammers from our engagement, from our comment section for each video.
Unfortunately, they continue to show up in every video. I'll continue to stay very diligent
in removing them each and every time, but I wanted to make sure you were aware that someone
is reaching out using my name and falsely representing themselves. All right, let's get
into tonight's video because, man, we had an interesting day today. I think there is a lot
of posturing, let's say, before tomorrow's CPI report gets dropped in the street because we saw
the entire memory sector down today, but it wasn't just memory. It was the entire tech sector. Take
a look at the NASDAQ 100 you can see on the screen and you can see everything from across compute
to memory to connectivity players to the picks and shovels were all down today. It was a risk
off kind of day. All of our indexes were down today. We saw the memory sector just digest some
of the gains that we've had over the last couple of days. We certainly have had some macro jitters
today across the tech sector, and I think there's good reason for it. Number one, I think we saw a
bit of a hotter than expected producer price index numbers that got dropped on the market today.
We saw it. We continue to see oil, Brent crude rise up over $100 a barrel. We're worried about
yields, but most importantly, I think we're seeing profit taking right now in anticipation
to the CPI report that will be dropped tomorrow on the street at 830 a.m. Eastern. So we have to
keep our eye on that. But I think this is the environment that we found ourselves in today,
and it was a risk off type of day for capital. It moved out of the memory players. It moved
out of tech stocks overall. And I think it was because people are anticipating potentially a
hot report tomorrow. And at the end of the video, I'm going to walk through two different types of
scenarios that we could see tomorrow with the CPI report and the impact that it'll have on the
memory sector. But even despite what you see on the screen today, all of the macro noise that you
see, there was significant physical data that came out today that just confirmed the hardware
ecosystem overall. And the structural bull case, it continues to signal stronger demand in the
market right now. There's three things that I want to walk through in tonight's video. One,
I want to talk about the demand signal we had today with TSMC's August numbers.
We'll go through those numbers and talk about what it actually means for the memory sector,
because we are seeing record historic numbers right now for TSMC. They beat July with their
August numbers, and July was the best numbers they've ever had. I want to talk about the
physical bottleneck that we see. So demand is strong, but we're continuing to see the physical
bottleneck in the market today. And it's not only impacting memory, it's actually spilling over into
the accelerators as well. It's spilling over into the compute players and the finished AI chips.
There was strong signs and signals for that out of China today we'll walk through. And finally,
let's talk about the architectural solution that SK Hynix presented today at their future forum
event when they're talking about their new roadmap for memory, the full stack AI memory roadmap. So
we'll talk about that together, what it means for the memory sector overall. That's what I want to
cover in today's video, those three big topics. But if you haven't done it yet, make sure you hit
like, make sure you subscribe to the channel. I want you to be part of Market Signal. You know
that. Let's continue to get our numbers higher. Let's continue to grow our base of subscribers
across YouTube and make sure that every investor that needs to hear this data that needs to get
this knowledge to ensure that they're a better investor going to get access to this content.
All right, guys, let's start with the TSMC data. Like I said, there was massive signals that came
out of TSMC today. Let me pull that up. You can see in DigiTimes here, it talks about the TSMC's
August revenue jumped 53% as AI chip demand drives record growth. If we scroll through some of the
data overall, you can see that their TSMC reported strong revenue growth in August, underscoring
continued momentum in AI related semiconductor demand. TSMC said that consolidated revenue
reached just over 16 billion at 16.3 billion in August. It was up 10% over July, which was a
record month for them. And it's up 53% year over year. And it's also, if you look at the first
eight months of the year, it is representing almost a 40% year over year for that first
eight month period. This is massive demand that the company is showing right now. And just take
a look at this graph here. You can see just the momentum they've had. Even from February,
we continue to have strong growth in March, a little bit of pullback in April, but then we see
continued sequential growth from May, June, July, and August, hitting record numbers for the company
overall. And when you combine what we saw from TSMC in July and August, they have already generated
over 31 billion. Their guidance was, I think, between 44 and 45 billion overall for the quarter.
So if you take a look, like I said, they already have achieved 31.1 billion. So for them to hit
the high end of their guidance, you can see right here in that paragraph here that they've guided
for the third quarter between 44.6 and 48.8 billion. So if you factor in July and August of
31.1, you have to realize that for September, TSMC only needs to hit about 14 to 15 billion
in revenue. So they could have a slight decrease in September over August and still easily blow
past Wall Street's expectations. I don't think they'll have a slowdown in September. The point
is that they are going to smash their third quarter estimate or the guidance they gave.
They're going to blow past that 45.8 billion overall just based on what we saw in July and
August and based on the other demand signals that we see in the market. We know this matters for
memory investors because TSMC does not, they don't operate in a vacuum. It sits directly underneath
NVIDIA, AMD, Broadcom, Marvell, and every other major hyperscaler custom ASIC program. A record
monthly revenue print of this magnitude, it completely dismantles the AI slowdown or the
CapEx cliff narrative that's in the marketplace. And more importantly, we know that there is a
strict one-to-one physical relationship between logic silicon and memory silicon. You cannot ship
a record volume of Blackwell GPUs, AMD Instinct accelerators, or custom hyperscaler XPUs without
placing an equivalent requirement of high bandwidth memory, high density server DRAM,
or enterprise flash storage right along with it. So TSMC's accelerating wafer output is the ultimate
leading indicator that the physical demand pipeline for Micron, SK Hynix, Samsung,
and SanDisk remains completely open right now in the marketplace. And I think that if you look at
TSMC, it proves that the foundry layer is pumping out logic at record speed. But when you attach
that much compute to the grid, you run headfirst into a brick wall that we know very well in this
channel, which is called the high bandwidth memory wall. And we received confirmation today,
if you pull this up, like I mentioned in the beginning, confirmation that memory is not only
impacting, it's not only impacting memory wall, it's impacting compute and it's impacting the
finished AI chips overall. So this confirmation today came out of China, Reuters reported it,
that major Chinese AI accelerators, including Huawei, as well as CambriaCon, are sharply
hiking up the selling prices of their finished AI chips because they simply cannot secure
HBM at reasonable costs. So let's walk through some of the data you can see in here. And it says
rising memory costs are adding pricing pressure across China's AI chip market, with major domestic
suppliers reportedly reporting raising prices for their latest products. We're seeing everything,
if you look at Huawei's product, they are increasing, the increases represent a 20 to 50%
increase from prices quoted to customers just two months ago. Huawei is not alone in raising
prices though. CambriaCon has also repriced its next generation chip, tentatively called the 690,
it's repriced at 20 to 30% above levels indicated just two months ago. And the reason for this is
because HBM shortage is driving up the Chinese AI chip costs. We know that HBM shortage is ripping
across the entire globe right now, but it is having a significant impact on China's AI industry. The
current administration, they triggered export controls on certain advanced HBM products to
China back in December of 2024. And Chinese chip makers have increasingly turned to gray market
channels to secure supplies. HBM obtained through these channels typically costs several times more
than what buyers pay, or buyers outside of China pay. So this is driving up their, and it's driving
up, their ability to get these advanced chips is hurting the Chinese ecosystem, the Chinese
memory. This is one of the cleanest real world confirmations of the physical memory wall that
we have seen yet. And today's story, it proves that the HBM scarcity, or this story proves that
the HBM scarcity has moved upstream. Memory is no longer just taking a bigger slice of the bill of
materials, it is actively setting the final retail price of the AI compute card itself. Just take a
look, we also know that China is trying desperately to expand their domestic HBM supply. CXMT has
begun small scale production of HBM 3E, and it plans to significantly expand its output into
2027. And they have CambriaCon, CambriaCon as well as Alibaba, they are reportedly testing this
memory for their processors. But until we see that, they're going to continue to experience
shortage, their ability to get their price on this advanced type of memory, and it's having an
impact on the finished product overall. So memory has officially shifted from a secondary component
to the ultimate structural control point in the global AI infrastructure. And when memory becomes
that availability, like for memory and becomes the primary bottleneck that is capping the global
compute output, we see the hardware architects, you know, they're not going to just sit on their
hands, they have to redesign the memory stack from the ground up. And it's exactly what we've
been talking about in this channel, how the architecture is changing. I'm going to share
with you what SK hynix presented, but know that micron presented something very similar,
it's on their website, if you look through in their investor presentations, their presentation,
they walk through actually, I actually will drop that link below for micron. And I'll drop the link
below in the description for this video for SK hynix as well. So you guys can see the exact
architecture that they're coming up with. But this article here in Seoul Economic Daily said
memory race is now a winding road, and it unveils an AI fab plan. We walk through it, there's a
couple key quotes that I wanted to call out in this article. What they said is that if memory,
the memory market of the past was a straight road where the contest was over, who could run the
fastest. Today, it is dry. It's like driving on a winding road that changes moment from moment.
Simply building faster, higher performing memory is no longer enough. Companies must read shifts
among customers and across the AI ecosystem and change how they work, how they run their
production floors, and their memory technology along with it. So this was a big presentation
today that came out from SK hynix in their their future forum 2026 future forum presentation,
they walked through three pieces of it overall. I'm not going to go through how they've actually
talked about how AI is transforming their production floor through their changing that
and creating something called an AI fab. But I do want to talk about what they're talking from a
memory perspective. So the shift alongside the AI shift on the production floor, SK hynix is
changing its memory strategy. As AI evolves beyond simple question answers into the agentic AI,
it sets its own plans and works over long stretches, the types and volumes of data
that must be stored and retrieved are growing. So SK hynix presented a full stack AI memory strategy.
The approach combines a range of memory from 3D stack DRAM, high bandwidth memory, high bandwidth
flash, according to the characteristics of AI workload. The plan is to expand it into a company,
expand themselves into a company that supplies memory architecture suited to the entire AI system.
And they must, they must evolve beyond a simple memory supplier into a full stack AI memory
creator. Let me show you the actual report that they went through. I just want to call out a
couple of the images from this. And like I said, I will post this specific form, this article in
the description if you're interested and want to read through it more. There's also a link for the
entire video of the presentation. But if I scroll down into the couple of the pieces here, the one
I want to get to is first is the shift itself, an evolution driving the shift in systems memory.
You can see on the left, it's the evolution from the past, which was the training or the
generative AI, where we kind of have question answer to the agentic AI from a plan and execute.
And there is a need for diversification of the AI state from a, you know, you look on the left,
into a session state, into a long-term state. And it's ultimately driving a memory hierarchy
driven by state diversification. You can see everything from HBM, DRAM, CXL, and SSD.
So memory usage matched to the AI state characteristics. If we go through further,
there's another image I wanted to show you here I really like is the AI data center pain points
and the memory solution. So you see the pain point on the left that SK Heinz is calling out,
this proliferation of agentic AI right now in the marketplace. We see the memory bottleneck as well,
and that, you know, the need for high capacity, high bandwidth required, and just the scarcity
of that resource that we have in the market. Well, their solution is this four-tier memory
pipeline where you have HBM as ultra-fast, DRAM as a high-performance general purpose,
CXL is the expendable, flexible kind of memory solution there. And finally, large capacity
persistent scale through SSD, and a new memory technology being added on. So that system level
optimization is what they're calling it, where you have memory tier times a heterogeneous
accelerator combination. It's going to maximize the AI infrastructure efficiency
through system-wide optimization. But you can clearly see when we take a look at what SK
Heineck's plans are, that they are moving to shift to meeting the customers where they need,
develop this roadmap, and look at a kind of future run for memory. This is a massive structural
takeaway for us long-term investors, because up till now, Wall Street evaluated the memory
bull run entirely on HBM wafer allocations. But what SK Heineck's proved today is that HBM,
or what they discussed, or they pointed out, is HBM is only one layer for the future architecture
of memory. Agentic workloads are expanding total addressable market across every single memory
format. It pulls high density NAND through HBM, through high bandwidth flash directly into an
active bandwidth layer. It's validating SanDisk's enterprise flash strategy while ensuring companies
like Micron, SanDisk, SK Heineck's can capture significantly more dollars per server across a
highly specialized multi-tiered architecture. So we bring it all together and kind of look at the
themes for today. We saw that TSMC proved demand at the foundry level is running at record highs
right now. August was over July, which was already a record month. August is the new record month.
Let's see what they can do in September. HBM's scarcity continues to be the theme across all
the markets, and it's proving that memory holds the ultimate pricing leverage over compute.
And finally, SK Heineck's full stack roadmap shows that memory stack is expanding into a
multi-layer growth engine. However, for tomorrow, as we enter tomorrow, we know that the trading
action will be dictated by the macro headlines, particularly the CPI report that gets dropped at
8.30 Eastern AM. And here's two scenarios that I think could play out, how we should read it,
or the impact that it could have in the memory sector overall. Here's the two scenarios that
I see. So if we come in with a CPI that looks hot, like you see on the left, I would expect
an immediate algorithmic selling at the opening bell, particularly across high data technology
and stocks and memory names as rate cut expectations get pushed back and treasury
yields may potentially spike. But the reality is here, guys, that this doesn't change anything that
we talked about. TSMC's record wafer shipments are still there. HBM pricing power is still in the
market. And for long-term investors, macro-driven dips and high-quality memory balance sheets have
historically offered prime accumulation windows driven by interest rate noise rather than
fundamental weakness. But if we do see what you see on the right, like a cooler report overall,
I think this could trigger a swift, short-covering rally in the market, easing rate fears and driving
capital right back into the leading AI infrastructure names. If the macro headwinds clear,
memory equities can quickly re-rate upward to price in the physical realities that we just
talked about that are in the market today, the physical realities around HBM, TSMC's output,
and what we're seeing from the memory manufacturers. We need to keep our eyes
kind of locked on the physical hardware data, the physical hardware data of this sector and
not just a single day trade or the tape that we saw from today, only that doesn't define the sector.
That what we saw over the last couple of weeks will continue. Let's pray that we have a good
CPI report for this sector tomorrow. We need it to come in cool so we can continue to get back on
a bull run like we've seen for the sector overall. But let me know your thoughts on today. Drop your
comments below. Let me know what you think we'll see with the CPI print tomorrow at 8.30. And make
sure you hit that like and subscribe button. And guys, I'll see you on the next one.