At CES 2026, AMD has unveiled its new MI440X artificial intelligence chip while outlining an ambitious roadmap for future processors, as the company steps up efforts to compete in the fast growing AI hardware market dominated by Nvidia.
The MI440X targets enterprise customers seeking high performance AI computing in existing data centers. Designed for compact eight GPU systems, the chip supports AI training, fine-tuning, and inference without requiring new facility builds. AMD said this approach helps organizations scale AI workloads more quickly and efficiently.
Alongside the launch, AMD introduced its Helios rack platform. The system combines 72 upcoming MI455X GPUs with EPYC “Venice” processors to deliver up to three exaflops of AI performance per rack, supported by 31 terabytes of high-bandwidth memory. The company positioned Helios as a solution for large-scale AI clusters.
AMD also highlighted a deepening partnership with OpenAI. OpenAI co-founder Greg Brockman announced plans to deploy six gigawatts of AMD AI compute, beginning with one gigawatt of MI450-series GPUs in the second half of 2026.
Looking ahead, AMD previewed its MI500 series GPUs, expected in 2027, promising major performance gains using advanced manufacturing and memory technologies. With customers including Meta, Microsoft, and Oracle, AMD aims to steadily narrow the gap in the AI accelerator market as deployments expand.
