Advanced Micro Devices (AMD -1.27%) has emerged as the no. 2 player in the AI accelerator market, although it remains far behind Nvidia. While Nvidia is now generating $30 billion per quarter in its data center segment, AMD has said that it expects to produce around $5 billion in all of 2024 from AI accelerator sales.
With demand for AI accelerators still booming, and with AMD’s market share small, there should be plenty of room for the company to grow its AI business quickly. However, 2025 could be less than stellar year for AMD if a new forecast from a Wall Street analyst is to be believed.
A potential growth slowdown
An analyst at Wolfe Research downgraded AMD stock on Thursday to the equivalent of “hold” and removed his price target entirely due to concerns about data center GPU growth. The analyst pointed to build plans from original design manufacturers to justify the increased pessimism.
The Wolfe analyst pegs AMD’s data center GPU revenue between $1.5 billion and $2 billion for the fourth quarter of 2024. For 2025, the forecast was lowered to just $7 billion, far below that $10 billion that analysts are generally expecting. That annual forecast basically predicts that AMD’s data center GPU revenue isn’t going to grow much quarter-to-quarter this year.
Of course, take any analyst opinion with a grain of salt. We’re less than three weeks into the new year, and the AI industry moves extremely quickly. AMD has some compelling AI accelerators on the market, and they’re clearly winning over customers. Oracle is using AMD’s MI300X GPUs for its cloud data centers, and Vultr, a smaller cloud provider, has chosen the same AMD GPU to build its AI offerings around.
One headwind AMD and all other non-Nvidia competitors face is the maturity of Nvidia’s software ecosystem. Nvidia GPUs support a proprietary software layer called CUDA, which has been around for a long time and is widely used across industry and academia for accelerated computing tasks. Even with capable hardware from AMD, the path of least resistance for those building out AI clusters is still Nvidia.
Software is a big reason why Intel’s efforts at breaking into the AI accelerator market have largely failed so far. Intel has an added complication that its Gaudi AI chips are not traditional GPUs, so the architecture also acts as a hurdle. AMD has fared much better than Intel, but it’s still far behind Nvidia.
New AI chips could help
The software situation should improve over time, and new chips from AMD could help boost sales this year if they’re well received by the market. The MI325X is expected to be widely available from system manufacturers in the first quarter, although that chip uses the same architecture as the previous MI300X.
The next generation Instinct MI350 family is expected to launch sometime in 2025, likely later in the year, and AMD is promising big performance improvements. Based on a new architecture, AMD is claiming that the MI350 family provides up to 35 times the AI inference performance compared to the MI300 family. The MI400 family will then come in 2026, built on another brand-new architecture.
Even with a quick pace of product launches, AMD is up against an equally quick Nvidia, and it will continue to struggle against its software disadvantage. While AMD’s AI chip sales this year are yet to be determined, any shortfall relative to expectations could send the stock tumbling.
