Prediction: This Artificial Intelligence (AI) Semiconductor Stock Will Outperform Nvidia Over the Next 3 Years

When it comes to artificial intelligence (AI) semiconductor stocks, Nvidia (NASDAQ: NVDA) is the gold standard, and there is still a lot to like about the stock. It remains the dominant chip company for AI model training, and that is unlikely to change. Most early foundational AI code was written using its CUDA software platform and optimized for its graphics processing units (GPUs), creating a wide moat in this arena.

Meanwhile, Nvidia hasn't rested on its laurels. The company is a huge player in data center networking, which has formed the base for it to become a complete AI infrastructure player offering end-to-end rack-scale solutions designed for specific AI tasks. The company also created its own custom Arm-based central processing units (CPUs), which will serve it well as the rise of agentic AI drives servers focused on AI agents to contain more CPUs.

Nvidia has also positioned itself well for the inference market through its "acquisition" of Groq and its language processing units (LPUs). LPUs contain a small amount of SRAM (static random-access memory) and are strung together to help reduce latency and speed up inference speeds. For systems designed for inference, Nvidia will combine its GPUs, which will handle the compute-heavy pre-fill phase of processing a prompt, with LPUs, which will provide low latency for the decode phase of generating a response.

Nvidia continues to produce incredible growth, with its fiscal first-quarter revenue soaring 85% to $81.6 billion while its adjusted EPS surged 140% to $1.87. Despite that, the stock is attractively valued, trading at a forward P/E of about 24.5 times fiscal 2027 (ending January 2027) analyst estimates. Given that, I think the stock still looks like a buy. However, given its size and the law of large numbers, I predict another AI semiconductor stock, Advanced Micro Devices (NASDAQ: AMD) can outperform it over the next three years.

Image source: The Motley Fool

AMD: Ready to take on Nvidia in inference

While Nvidia controls the market for large language model (LLM) training, AMD is set to be a serious competitor when it comes to inference. Importantly, this is the faster-growing market and also expected to become the much larger one. According to projections from Bloomberg Intelligence, the inference market is projected to grow at a 32% compound annual growth rate through 2032 and become nearly double the size of the AI model training market, hitting annual spending of $1.3 billion.

Inference is less technically demanding than AI model training, and AMD has done a great job improving its ROCm software stack over the past few years. Meanwhile, inference is generally more memory-bound than compute-constrained, and the company's chiplet design, which can be packaged with more memory and is designed to act as part of an entire system, reduces latency.

AMD also made two acquisitions in the space to better position it for the inference market. First, it acquired memory optimization company MEXT, which offloads less-frequently used data from DRAM to underused flash memory and then uses predictive AI to transfer it back before it's needed. It then bought inference chip start-up Taalas, which hardwires AI models directly onto its chips to bolster inference performance. It uses specific AI models, so it isn't as flexible, but it's a cheap and fast inference alternative.

Along the same lines, AMD also partnered with Cerebras to offer a disaggregated system for inference. AMD's Helios system will deal with the pre-fill phase of inference, while Cerebras' ultrafast but expensive solution will handle the decode phase. It is a partnership that should benefit both companies.

At the same time, AMD also has a huge opportunity with agentic AI. The company is a leader in server CPUs, and has been steadily taking share in this market away from Intel. With the ratio of GPUs to CPUs moving from 8:1 for training to 1:1 for agentic AI, AMD sees this becoming a $220 billion market in the next few years.

As a much smaller company than Nvidia, AMD really has the opportunity to see explosive growth and for its stock to outperform over the next three years as it makes serious inroads in the inference and agentic AI markets. With big deals already in place and its revenue growth about to take off, I think the stock can outperform Nvidia over the next three years.

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