AMD is trying to make a dent in the AI graphics card market, but will it be able to eat into Nvidia’s market share?
The market for artificial intelligence (AI) accelerators is currently dominated by Nvidia (NVDA -4.69%), which has built a technology lead over rivals and controls an estimated 85%-plus share of this space.
However, fellow chipmakers, such as Advanced Micro Devices (AMD -5.22%) and Intel, are also trying to corner a piece of this lucrative market. AMD recently unveiled a new AI chip — the Instinct MI325X — claiming that it’s better than Nvidia’s current flagship H200 processor. However, shares of AMD sank following the announcement.
Read on to discover why that may have been the case and check if AMD’s latest AI chip should be a cause for concern for Nvidia investors.
AMD’s MI325X seems impressive, but maybe a bit too late
AMD points out that its MI325X data center accelerator is equipped with 256 gigabytes (GB) of HBM3E high-bandwidth memory that supports a bandwidth of 6 terabytes (TB) per second. According to AMD, its latest chip has 1.8x more capacity and 1.3x more bandwidth than Nvidia’s flagship H200 processor, which is based on the Hopper architecture.
As a result, AMD claims that the MI325X offers 1.3x greater compute performance, theoretically, along with an identical advantage in AI inference on the Llama 3.1 and Mistral 7B large language models (LLMs). The company is on track to start the production of the MI325X from the current quarter and says it will be widely available in the first quarter of 2025.
Given that Nvidia’s H200 processor offers 141GB of HBM3E memory at a bandwidth of 4.8TB/second, there’s no doubt that AMD’s new AI chip is superior on paper. But at the same time, investors should note that the H200 was announced nearly a year ago and started shipping in the second quarter of 2024. AMD’s chip, therefore, is around nine months later to the market than the H200.
On the other hand, Nvidia is set to level up its game in the AI chip market with its next-generation Blackwell processors. The upcoming B200 Blackwell graphics processing unit (GPU) is set to pack 208 billion transistors, as compared to the 153 billion transistors on the MI325X. More importantly, the B200 is reportedly manufactured using Taiwan Semiconductor Manufacturing‘s 4-nanometer (nm) 4NP process node, compared to the 5-nm process node for the MI325X.
Nvidia will be packing a much higher transistor count into a much smaller chip, which means that its B200 processor is likely to have more computing power and be more power-efficient at the same time. Also, the B200 is reportedly going to offer a higher memory bandwidth of 8TB/second. The good part is that Nvidia expects to “get several billion dollars in Blackwell revenue” in the fourth quarter of its ongoing fiscal year, which will run from November 2024 to January 2025.
What this means is that Nvidia would have moved ahead of AMD by one generation by the time the latter’s latest chip is widely available to purchase. As a result, there doesn’t seem to be a strong reason for Nvidia investors to be wary of AMD’s latest offerings. However, there does appear to be a silver lining for AMD investors.
The new chip may not beat Nvidia but could give AMD a nice boost
Nvidia is running away with the AI chip market and seems on course to clock almost $100 billion in data center revenue in the current fiscal year. AMD, on the other hand, is expecting its 2024 AI data center GPU revenue to hit $4.5 billion, which pales in comparison to Nvidia’s potential revenue from this segment.
However, AMD doesn’t need to beat Nvidia in AI GPUs to supercharge its growth. It simply needs to become the second-biggest player in this market, which AMD believes could be worth a whopping $500 billion in 2028. The company started selling its AI GPUs toward the end of 2023 and sold $6.5 billion worth of data center chips last year, including its Epyc server processors.
This year, the company has generated $5.2 billion in data center revenue already in the first six months of 2024, which is nearly double its data center revenue in the same period last year. At this run rate, AMD could generate nearly $10.5 billion in data center revenue this year, $4.5 billion of which will come from sales of AI GPUs. Assuming AMD could capture even 10% of the AI GPU market in 2028, it could generate $50 billion in annual revenue from this segment, which would be a big jump from this year’s projections.
Additionally, AMD’s MI325X processor could witness robust demand, as Nvidia pointed out on its previous earnings conference call that its H200 processors are expected to keep selling nicely despite the arrival of the new Blackwell chips. More specifically, Nvidia estimates that shipments of Hopper-based chips, such as the H200, are set to rise in the second half of the current fiscal year, suggesting that there’s space for the MI325X, despite the arrival of more powerful processors.
Moreover, AMD’s growing presence in the AI GPU market is expected to boost its growth from next year. While the company’s revenue is projected to jump 13% in 2024 to $25.6 billion, the forecast for the next couple of years appears to be even better.
AMD Revenue Estimates for Next Fiscal Year data by YCharts.
Even if AMD is unable to beat Nvidia in the AI chip market, it could turn out to be a solid AI stock in the long run if it manages to carve out a small niche for itself. At the same time, investors should note that AMD can capitalize on other AI-related applications, as well in the form of AI-enabled PCs and server processors.
Even though AMD stock has delivered tepid gains of just 14% in 2024, investors would do well to keep it on their watch lists as it could soar impressively, thanks to the huge opportunity in AI GPUs and other markets.
Harsh Chauhan has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Advanced Micro Devices, Nvidia, and Taiwan Semiconductor Manufacturing. The Motley Fool recommends Intel and recommends the following options: short November 2024 $24 calls on Intel. The Motley Fool has a disclosure policy.
