In the competitive field of AI graphics cards, Advanced Micro Devices (AMD) is endeavoring to make an impact and capture a portion of the market predominantly controlled by Nvidia. Nvidia presently maintains a significant lead in the AI accelerator market, commanding over an 85% share due to its technological advancements.
AMD, alongside other chipmakers like Intel, is attempting to gain traction in this viable sector. Recently, AMD announced its latest AI chip, the Instinct MI325X, which the company claims surpasses Nvidia’s leading H200 processor. Despite this announcement, AMD’s share prices experienced a decline.
The MI325X boasts impressive specifications, with 256 gigabytes of HBM3E high-bandwidth memory and a bandwidth capacity of 6 terabytes per second. AMD states that this provides 1.8 times the capacity and 1.3 times the bandwidth of Nvidia’s top H200 processor, resulting in 1.3 times greater compute performance and a similar advantage in AI inference on large language models like Llama 3.1 and Mistral 7B. Production of the MI325X is set to begin in the current quarter with wide availability expected in the first quarter of 2025.
Although the MI325X appears superior in comparison to the H200 processor, which features 141GB of HBM3E memory and a bandwidth of 4.8TB per second, it enters the market almost nine months after the H200. Concurrently, Nvidia is preparing to introduce its next-generation Blackwell processors. The upcoming B200 Blackwell GPU, reportedly consisting of 208 billion transistors and manufactured using Taiwan Semiconductor Manufacturing’s 4-nanometer process, surpasses AMD’s 5-nanometer MI325X chip. Nvidia anticipates this next-generation B200 processor to deliver enhanced computing power and efficiency alongside an improved memory bandwidth of 8TB per second. Nvidia forecasts several billion dollars in revenue from Blackwell processors by the end of its fiscal fourth quarter, covering November 2024 to January 2025.
As Nvidia progresses with its next-generation technology, stakeholders might remain confident in Nvidia’s leadership, yet AMD’s developments could present notable opportunities for growth. Nvidia is poised to potentially achieve close to $100 billion in data center revenue within the present fiscal year. Conversely, AMD projects its AI data center GPU revenue will reach $4.5 billion in 2024, which, while significant, remains a fraction of Nvidia’s expected intake. However, it should be noted that AMD aims to establish itself as the second-leading player in a market predicted to reach $500 billion by 2028.
Despite not overtaking Nvidia, AMD’s involvement is noteworthy. With ongoing substantial data center revenue growth, AMD projects a 13% rise in overall revenue to $25.6 billion for 2024, with subsequent years looking even more promising. Additionally, demand for AMD’s MI325X may remain strong, as indicated by Nvidia’s continued anticipated sales of its H200 processors despite the forthcoming Blackwell chips.
Investors may find AMD has potential as a promising AI stock, given its prospects in AI GPUs and other applications, including AI-enabled PCs and server processors. Though AMD’s stock performance in 2024 has been modest with a 14% increase, the potential for significant growth remains, given the vast opportunities in the AI sector.
