This year, it’s all about inferencing. The recent stock market dip can partly be attributed to what #deepseek announced, but there are also other players in the training and inferencing space. This chart highlights how memory sizes differ for @AMD and @Nvidia’s latest offerings. If this were a speed chart, the GB200 would be on top. The case in point is that we, as developers, have become fatigued by the constant announcements of new processors. Processors are merely a means to an end. Training on one processor doesn’t necessarily mean the model will perform well on another. Focusing on a single chip as the “be-all and end-all” of AI development doesn’t help the broader narrative (though it may boost a company’s stock price). Developers will tell you that other processors, such as @Intel’s Gaudi 2 and the newly released Gaudi 3, can do the job, depending on the application. For example, AMD’s MI325X, as shown in the diagram below, is better suited for genome sequencing due to its memory advantages. Meanwhil