AMD unveiled super sophisticated AI chips at an event held in San Jose, California, for large-scale artificial intelligence endeavors. AMD CEO Lisa Su unveiled the Instinct MI400 series, which will start shipping in 2026. Open AI CEO Sam Altman joined the event, announcing that Open AI will soon integrate AMD’s latest bespoke AI chips deep into its sprawling operations. Altman expressed enthusiastic support for the new hardware, highlighting its immense potential to significantly accelerate AI advancements.
Instinct MI400 GPUs are engineered for hefty AI workloads on large scales and integrate into entire racks of servers seamlessly. These GPUs boast 432GB HBM4 memory and deliver 40 petaflops FP4 performance alongside a remarkably high 300 GB/s scale-out bandwidth for industry-leading rack-scale AI. They’re finely tuned for cranking out humongous AI models and slinging around distributed inference effectively at massive scales.
Lisa Su underscored AMD’s AI chips being leveraged by major tech behemoths like Meta and Microsoft alongside OpenAI, Tesla, Oracle, and xAI already. AMD aims to introduce new AI chips each year, focusing on delivering affordable and scalable solutions to drive the adoption of large-scale AI systems by 2028.
Nvidia presently dominates the data center GPU market, largely due to early innovation with CUDA, a foundational AI software stack. Nvidia’s Blackwell chips support large configurations with as many as 72 GPUs for humongous AI systems. AMD’s MI400 series boldly challenges Nvidia’s stranglehold on AI processing by introducing competitive solutions at rack scale.
AMD’s stock surged 9% after the announcement, with analysts keenly anticipating a strong recovery in the company’s GPU business.
