I've Been Wrong About This Tech Stock for Years, but I've Finally Bought Shares - The Globe and Mail

One of the most important lessons an investor can learn is to be adaptable. It's easy to get caught up in a thesis or mindset and want to stick with it no matter what. However, it is important to be flexible as companies and industries evolve and new information presents itself.

This holds true for both buying and selling. When the information changes, be ready to reevaluate and change with it. One stock I've done this with recently is Advanced Micro Devices(NASDAQ: AMD). For the past few years, I thought of AMD as a distant No. 2 player in the graphics processing unit (GPU) market that was unlikely to ever become an important player in the artificial intelligence (AI) infrastructure build-out.

However, a lot has changed over the past six months. Let's see how -- and what this means for investors.

Riding powerful trends

AMD has always been stuck behind Nvidia and has faced a steep uphill climb to compete in the GPU market due to its subpar software ecosystem that often made its GPUs unusable out of the box for AI model training. However, after more than two years of investment, the company has vastly improved its ROCm software platform, while the shift of software developers to open-source AI frameworks has helped open the door, especially in inference. Importantly, inference is expected to eventually become a much bigger market than large language model (LLM) training over time.

This helped the company land two major hyperscaler (owners of large data centers) customers for its GPUs over the past several months. Both OpenAI and Meta Platforms have agreed to commitments of 6 gigawatts worth of chips in the coming years, estimated to be worth more than $100 billion. While AMD provided both companies with warrants worth up to 10% of the company, based on deliveries and AMD stock hitting certain price triggers, it got the company into two of the biggest AI infrastructure spenders and ensured they would need to implement ROCm within their ecosystems. Given the potential size of the inference market, this trade-off looks like a smart move.

AMD's GPUs, meanwhile, are well positioned for the inference market. The company uses a modular chiplet design that packs in more memory capacity. Inference is often more memory-bound than compute-bound, meaning performance is limited by how quickly data can move from memory to the processor, not how powerful the GPU is. As such, AMD's new MI450 series, which has 1.5 times the memory capacity of Nvidia's upcoming series of Rubin chips, looks like it has an advantage in this area, at least on paper.

At the same time, another new powerful trend began to work in AMD's favor. AMD had long been taking market share from Intel in the data center central processing (CPU) market. This was a nice growth driver, but overshadowed by GPUs, as not only are GPUs much more expensive, but with AI infrastructure, they would typically be deployed at an 8-to-1 GPU-to-CPU ratio. With servers dedicated to inference, that ratio would be cut to around 4-to-1.

Image source: The Motley Fool.

However, with the rise of agentic AI, the high-performance CPU market looks set to explode. While GPUs are great at providing the muscle needed for fast processing power, AI agents need chips that can handle sequential reasoning and work with other tools. That is where CPUs come in, as they work as the brains of the operation. With agentic AI, the GPU-to-CPU ratio is expected to move to around 1-to-1 or even slightly favor CPUs. Meanwhile, AI agents are best served by high-performance CPUs with high core counts, as each core acts as an individual workplace to perform tasks. The more cores, the higher the price of the CPU. AMD and Arm Holdings have said they see the data center CPU market growing to between $100 billion and $120 billion over the next five years.

The story with AMD has changed. It's gone from an AI infrastructure afterthought to riding two powerful trends. As such, I changed my long-held belief about the stock and now own shares, as I expect to see explosive growth from the company in the coming years.

Geoffrey Seiler has positions in Advanced Micro Devices and Meta Platforms. The Motley Fool has positions in and recommends Advanced Micro Devices, Intel, Meta Platforms, and Nvidia. The Motley Fool has a disclosure policy.