Here is my AMD Investment Thesis!
1/ $Advanced Micro Devices(AMD)$ has just signed a $100 billion chip deal with OpenAI.
This is just the beginning.
It has a path to become a $1 trillion company as it's catching up with $NVIDIA(NVDA)$ in both hardware and software.
Here is my AMD investment thesis: 🧵
Lisa Su holding two semiconductor chips in her hands. She stands in front of a large AMD logo on a dark background.
2/ AI infrastructure spending is accelerating.
Most people still don't see what's coming.
Jensen Huang thinks that annual AI-chip spending will reach $1 trillion by the end of this decade.
This is a massive total addressable market for both AMD and NVDA.
3/ NVDA has taken the lion's share of this market so far.
This was because of two reasons:
- It was the first mover in data center GPUs.
- Its programming platform, CUDA, was the standard.
Competitors were behind NVDA in hardware performance and software capabilities.
This is about to change as heavy AI workloads shift from training to inference.
Let me explain:
Image
4/ AI model training requires two important optimizations:
- Strictly directing the model through programming.
- Moving large amounts of data between servers.
NVDA has two critical products for this: CUDA and NVLink.
CUDA was the industry standard in parallel programming, while NVLink was the best-in-class server-to-server interconnect in the market.
This is why NVDA dominated the market while AI labs mainly competed in training better models.
This is about to change.
Image
5/ Gains from pre-training the models are flatlining.
Even though the available computing power is growing, the total data pool isn't growing as fast.
Thus, longer training using better chips won't likely result in substantially better models.
Once this happens, the competition will shift from training to inference.
Capabilities required to dominate inference are different than training, and AMD has the upper hand there.
Image
6/ Inference is about serving one request at a time as quickly and accurately as possible.
This requires keeping as large a part of a model as possible in a single server.
Memory is the most important capacity for this task.
AMD has a critical design expertise here: Chiplets.
Traditionally, different components of a chip, like memory and cores, were put in a single silicon die.
Thus, all the components needed to scale down together as chips got smaller, although memory is a less complex product and doesn't need to scale down that much.
This increased the cost and complexity.
AMD was the first big chip company to pivot away from monolithic design to chiplet-based architecture.
Image
7/ Chiplet design is based on manufacturing different parts of a chip on different nodes and connecting them together with a high-speed interconnect.
This allows designers to use lower-cost processes like 5nm for less complex memory parts while using more expensive 3nm processes for cores.
AMD has developed the best-in-class interconnect, Infinity Fabric, to connect different parts together at high speed.
This allowed AMD to pack more memory in a single chip while keeping the costs down, making it a perfect fit for inference workloads.
This is how MI355X offers considerably higher memory than Blackwell at a way lower cost.
Image
9/ OpenAI's massive 6GW AMD GPU order further substantiates AMD's advantage in inference.
As the gains from training further flatline, competition will shift to inference, and $AMD will actively win market share from NVDA.
This deal can grow further in the future as OpenAI actively participated in the development of AMD's MI450 series, as Sam Altman said:
10/ Valuation can still be attractive.
OpenAI's massive order validates the $AMD thesis and positions it to win market share.
McKinsey forecasts that the AI accelerator market will reach $500 billion by 2030.
Even if AMD reaches just 20% market share, it'll generate $100 billion in revenue.
Assuming it'll have a bit lower margins than NVDA, around 40%, it'll generate $40 billion in net income.
Assuming a conservative 25 times exit multiple, we get a $1 trillion company.
It's currently valued at just $360 billion.
Image
For whom haven't open CBA can know more from below:
🏦 Open a CBA today and enjoy privileges of up to SGD 20,000 in trading limit with 0 commission. Trade SG, HK, US stocks as well as ETFs unlimitedly!
Find out more here：
Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.
- AuntieAaA·10-08GoodLikeReport
- 闪电侠08·10-07OkkkLikeReport