Graphic processing units (GPUs) serve as a key component in the foundation of the world's artificial intelligence (AI) infrastructure build-out. Training AI models and running AI inference demands high-speed processing power, and it creates computational workloads that can best be handled using parallel processing. That positions GPUs -- originally developed to speed up the rendering of video game graphics -- as some of the best hardware options for providing the right type of processing power.
Two companies dominate the GPU market. Nvidia (NVDA -2.09%) is by far the leader in the space. Rival Advanced Micro Devices (AMD 0.10%), meanwhile, is trying to take Nvidia on and gain a meaningful share of the data center market. In 2024, Nvidia's stock has been the clear winner, trading up by about 175% as of this writing. AMD stock, meanwhile, is down by about 15% on the year.
But which stock is likely to perform better in 2025?
The GPU market keeps growing
The outlook for the GPU market continues to look strong. Most large hyperscalers (companies that operate mega-sized data centers) have already indicated that they plan to increase their AI-related capital expenditures in 2025. Demand for cloud computing is soaring due to AI, and all the major cloud computing companies are working to expand their infrastructure to meet that demand.
Meanwhile, GPU clusters are growing larger and larger as big tech companies and well-funded start-ups such as OpenAI and Elon Musk-owned xAI race to develop more sophisticated AI models.
It is expected that Meta Platforms will use 160,000 GPUs to train its upcoming Llama 4 model -- 10 times as many as it used for Llama 3. The case is similar with xAi's Grok 3 model, which is expected to require 100,000, up from 20,000 used to train Grok 2. Meanwhile, there is talk of companies deploying clusters with as many as 1 million GPUs in the near future.
Nvidia has been the biggest beneficiary of these companies' insatiable demand for GPUs, and its revenue has far outpaced AMD's. A big reason for this is that Nvidia long ago developed its free (but proprietary) CUDA software platform, which allows developers to program the GPUs they buy for tasks other than graphics rendering. As such, CUDA became the software platform on which many developers learned to program GPUs, creating a wide moat for the company. In the years since, Nvidia has added developer tools and AI-specific microlibraries through CUDA X, extending its software lead.
Today, AMD has its own GPU software platform, and it makes GPUs that are just as powerful as Nvidia's, if not more so -- at least, on paper. However, testing by independent research and analysis company SemiAnalysis has concluded that AMD's software is holding back the performance of its GPUs. In its report, it called AMD's out-of-the-box experience "unusable," and said that it needed "multiple teams of AMD engineers" to help it fix software bugs. In the end, the paper specs for AMD's latest GPU did not match its real-world performance. By contrast, SemiAnalysis described the out-of-the-box performance of Nvidia's H100 and H200 GPUs as "amazing."
This helps explain why Nvidia generated data center revenue of $30.8 billion last quarter compared to only $3.5 billion for AMD. Notably, though, both showed similar rates of data center revenue growth: Nvidia's rose by 112% and AMD's rose by 122%. Of course, Nvidia was doubling its revenue from a much larger base, which makes its performance all the more impressive.
One area in which AMD has been able to carve out a niche for itself is AI inference. SemiAnalysis noted that AMD's customers tend to deploy its GPUs in inference, which presents narrow, well-defined use cases. As such, one way AMD could start to gain market share in the coming years is if more of the GPU market shifts from training to inference. AMD's GPUs are cheaper than Nvidia's, so such a shift could potentially benefit it.
Image source: Getty Images
Valuation and verdict
From a valuation perspective, AMD is a modestly cheaper stock, trading at a forward price-to-earnings (P/E) ratio of 24 compared to nearly 31 for Nvidia. However, Nvidia has been growing its total revenue more rapidly (94% vs. 18% last quarter) as at AMD, GPUs are just one piece of a larger hardware portfolio.
AMD PE Ratio (Forward 1y) data by YCharts.
At this point, I think AI training will remain important over the next several years as companies continue to try to develop more advanced AI models. As such, I much prefer Nvidia's stock as a holding from here and think investors could still profitably add it to their portfolios in 2025.
