On Thursday, AMD unveiled its new artificial intelligence chip, the Instinct MI325X, aimed directly at competing with Nvidia's data center GPUs. Production of the MI325X is set to begin by late 2024. If successful, AMD's AI chips could challenge Nvidia’s pricing, which has benefitted from high demand and gross margins around 75%. The need for large data centers filled with GPUs to support advanced generative AI, such as OpenAI's ChatGPT, has spurred demand for alternative AI chips. Nvidia has largely dominated the data center GPU market, with AMD in a distant second place. AMD aims to capture a significant portion of the projected $500 billion AI chip market by 2028. CEO Lisa Su emphasized that AI demand is exceeding expectations and is driving investment growth. While AMD did not announce new major partnerships at the launch, it has existing collaborations with companies like Meta and Microsoft.

Specific pricing for the MI325X remains undisclosed, as it is typically sold as part of a server package. AMD will release new chips annually, with the MI350 planned for 2025 and the MI400 for 2026, in an effort to keep pace with Nvidia, especially as Nvidia prepares to launch its Blackwell chips next year. AMD's stock fell 4% following the announcement, while Nvidia's rose by about 1%. A major hurdle for AMD is Nvidia’s proprietary CUDA programming language, which locks developers into its ecosystem. AMD is enhancing its ROCm software to facilitate the transition of AI models to its chips, which it markets as accelerators. AMD claims the MI325X offers up to 40% better inference performance than Nvidia’s H200 chip for certain applications like Meta's Llama AI model. In addition, AMD announced new data center CPUs, EPYC 5th Gen, to bolster its CPU market share, which stands at 34%, still trailing Intel. These CPUs are designed to optimize data input for AI workloads, emphasizing the importance of CPU performance in AI applications, as most GPUs require a compatible CPU to operate.