Meet the Market's Newest $1 Trillion Stock: 3 Things to Know Before Buying AMD
I'm LongbridgeAI, I can summarize articles.AMD became the newest $1 trillion stock, driven by an AI boom and a 187% year-to-date rise. Key insights include: 1) AMD is positioning for growth in the inference market, leveraging chiplet designs and acquisitions like Taalas and MEXT. 2) Major GPU deals with OpenAI, Meta, and Anthropic total over $100 billion but involve issuing warrants and equity stakes. 3) AMD remains a leader in data center CPUs, targeting agentic AI growth to capture significant market share from Intel.
The artificial intelligence (AI) boom helped mint its newest $1 trillion company, with Advanced Micro Devices (AMD +0.34%) crossing the threshold on Sept. 21. The semiconductor stock had risen 187% year to date to reach a $1 trillion market capitalization.
Here are three things investors need to know about AMD before investing.
1. A big inference opportunity ahead
While AMD has long played second fiddle to rival graphics processing unit (GPU) maker Nvidia in the AI model training market, the company has positioned itself to be a winner in the inference market. This is a good thing, as Bloomberg Intelligence predicts that the inference market will become roughly double the size of the training market by 2032.
Inference is much more about quick access to memory than the raw compute power needed to train AI models. The company's GPUs can be packaged with more high bandwidth memory (HBM) due to their chiplet design, making them a solid choice for running inference. Inference has lower barriers to entry than training AI models, and the company's ROCm software platform has come a long way in the past few years.
Importantly, the industry's move to open-source frameworks has also better positioned AMD's GPUs for inference, as users no longer need to write custom code for their chips or rely on a previously buggy CUDA translation layer.
NASDAQ: AMD
Key Data Points
On top of that, AMD has now developed an entire end-to-end server called Helios, designed specifically for AI workloads, following its acquisition of ZT Systems. It also bought two memory-related companies that speak to its future server strategy.
First, it acquired memory optimization company MEXT, which has a predictive AI technology that can offload data from HBM to flash memory and then transfer it back before it is even requested. Next, it acquired the inference chip company Taalas, which hardwires specific models into its chips to increase inference speed. AMD also formed a partnership with Cerebras, combining its solutions with Cerebras' high-end system, which handles the decode phase of inference, while Helios handles the pre-fill phase, giving it a premium system at a lower cost.
2. AMD has big GPU deals in place, but at a cost
To push into the inference market, AMD forged large GPU deals with both OpenAI and Meta Platforms earlier this year. AMD will provide both companies with up to 6 gigawatts of its latest GPUs in deals estimated at over $100 billion each. Shipments for the first gigawatt are expected to begin in the second half of this year.
However, both deals also came at a price. AMD will issue warrants to both companies for up to 160 million of its shares, which is about a 10% stake in the company. Each tranche of warrants will vest based on AMD's stock price, up to the point where it hits $600, which it has already done, and on "other technical and commercial considerations."
AMD later struck a 2-gigawatt deal with Anthropic. In exchange, it will take a $5 billion stake in the AI model maker. It also has a smaller deal in place with Microsoft.
While its two GPU deals came at a price, they did ultimately open the door for the company in this lucrative market.
Image source: The Motley Fool.
3. A CPU leader
In addition to its big GPU opportunity, AMD is also a leader in the data center central processing unit (CPU) space. This market is booming with the rise of agentic AI, as servers dedicated to AI agents require significantly more CPU capacity. While AI training generally has a GPU-to-CPU ratio of 8:1, it drops to 4:1 for inference and to 1:1 for agentic AI.
As a result, the company sees this growing to be a $220 billion market over the next few years. AMD has consistently been taking share from Intel in this market and believes it can reach about 50% market share. Meanwhile, it has already developed new high-core CPUs designed for agentic AI.
Ultimately, this positions the company to chase two large growth opportunities.