AMD Unveils MI350 Series, Expands AI Market Reach

At its annual Advancing AI event, AMD launched the Instinct MI350 Series accelerators, including the MI350X and liquid-cooled MI355X, featuring advanced 3D chiplet designs and 288GB of HBM3E memory—60% more than Nvidia’s latest. The MI350 delivers up to four times faster AI compute and 35 times faster inferencing than the previous generation, supporting AI models with up to 520 billion parameters. Major deployments include Oracle's 27,000-GPU cluster, and over 60 organizations are now AMD customers, including seven of the ten largest AI companies. AMD also showcased an open, rack-scale AI infrastructure and previewed its Helios rack and MI400 Series chips coming in 2026, outlining a roadmap through 2027. While AMD still trails Nvidia in some aspects, it is closing the performance gap and emphasizing openness and price/performance benefits. These moves aim to expand AMD's presence in the AI and data center markets, targeting customers beyond Nvidia’s typical base.