AMD CEO Lisa Su unveiled the company's latest AI strategy and groundbreaking new chips at the Advancing AI 2025 event held in San Jose. The event showcased AMD's commitment to innovation in artificial intelligence, highlighting advancements in both hardware and software.
AMD's Vision for the Future of AI
Su emphasized AMD's relentless focus on making AI development easier for developers, with out-of-the-box capabilities, simplified setup, and increased community engagement. AMD has accelerated its release cadence, now shipping new features and optimizations every two weeks. Key models like Llama and DeepSeek are supported from day one.
Introducing the Instinct MI350 Series
A key highlight of the event was the launch of the AMD Instinct MI350 series, including the MI350X and MI355X. These new HPC/AI GPUs are based on a 3nm process node with up to 185 billion transistors. The MI355X boasts significantly faster performance, outpacing even the MI300 by a substantial margin. Partners like Oracle, Dell Technologies, HPE, Cisco, and Asus are expected to launch products featuring these chips in Q3 2025.
- MI350X: Equipped with HBM3e memory and support for FP6 and FP4 AI data types.
- MI355X: Offers impressive specs, including 8 TB/s aggregate memory bandwidth, 79 TFLOPs of FP64, and 20 PFLOPs of FP6/FP4 compute.
Software's Crucial Role
AMD also underscored the importance of software in unlocking the full potential of AI, with Vamsi Boppana, SVP of AI at AMD, discussing ROCm, AMD's software platform for GPU computing. The company recognizes that seamless integration between hardware and software is vital for driving AI innovation.
Competitive Edge
Analysts anticipate the MI355X AI GPU could come with more high-bandwidth memory and a lower price point than rival Nvidia’s B200 chip. AMD could also announce new high-profile customers like Amazon or OpenAI, further solidifying its position in the AI market. AMD’s advancements position it as a strong competitor against Nvidia.