Since the start of the artificial intelligence (AI) build-out, I viewed Advanced Micro Devices (AMD 0.03%) mostly as an afterthought, a company destined to just live deep in Nvidia's (NVDA +1.67%) shadow. The company was the No. 2 player in the graphics processing unit (GPU) space, but the gap was so wide between it and Nvidia that it felt it would never make any serious inroads.
After all, when it came to AI, Nvidia had actually outmaneuvered AMD a long time ago, way before AI became mainstream. AMD got into the GPU game in 2006 when it acquired ATI Technologies. While AMD was busy integrating ATI into its business, Nvidia released its CUDA software platform to allow developers to easily program its chips beyond their original purpose of speeding up graphics rendering in video games. It gave the software program away for free and embedded into places that were doing early research on AI. This led to most foundational AI code being written on its software programmed for its chips.
Image source: The Motley Fool.
AMD did not release its competing ROCm software platform until a decade later, at which time it was very far behind. Meanwhile, as the AI race heated up, ROCm became a liability, as the software platform was generally viewed as buggy and unusable out of the box. Almost all code was still written in CUDA, and converting it to ROCm was a pain. However, more recently, more programming has started to be written on newer open-source AI frameworks, like OpenAI's Triton, which has opened the door for AMD and a now much-improved ROCm.
With this opening, AMD was able to strike two large and important GPU partnerships with OpenAI and Meta Platforms. Both deals were structured in the same way, with both companies committing to 6 gigawatts of GPU capacity. In exchange, AMD also issued both customers warrants worth up to 10% of the company based on deliveries and its stock price. While these agreements came at a price, their size requires both companies to integrate ROCm into their data center ecosystems and incentivizes them to support AMD.
That's a big win for AMD, especially as the market begins to shift more toward inference. Inference isn't as technically demanding as large language model (LLM) training, so Nvidia's CUDA advantage isn't as formidable. Meanwhile, AMD said it now rarely gets requests to convert code from CUDA, as most of its inference customers today use other frameworks, such as vLLM or SGLang. While the total cost of ownership is the most important consideration with inference, with its GPUs considerably cheaper than those from Nvidia, this can really open the door for it to take some share if it can continue to close the performance gap.
NASDAQ: AMD
Key Data Points
The agentic AI catalyst
While the GPU market opportunity for AMD is starting to materialize, the company is seeing another big growth driver emerge in agentic AI. AI is moving beyond learning and content creation to action, and with it, AI agents need chips that can handle reasoning and work with other tools. That's not what GPUs do well, as they essentially provide the muscle. Instead, to handle these operations, data centers are going to need a lot more central processing units (CPUs).
That's a huge opportunity for AMD, as it's the current leader in data center CPUs. This market is expected to explode, with the ratio of CPUs to GPUs in AI data centers is expected to narrow considerably. The CPU market is already starting to become undersupplied, helping AMD raise prices. Meanwhile, the company is already designing its newest CPU architecture specifically with agentic AI in mind. Not to be overlooked, AMD's acquisition of ZT Systems will allow it to sell pre-configured racks designed specifically for agentic AI.
From afterthought to AI winner
AMD is no longer just a secondary AI player along for the ride. It has repositioned itself for both inference and agentic AI, setting itself up for very strong growth in the years ahead. The AI supercycle has room for more companies than just Nvidia, and one of those companies is AMD. That's why I recently bought AMD shares.
