AMD Unveils New Ai Hardware, Software, Infrastructure and Developer Cloud

AMD has outlined its expanding role in Ai infrastructure, revealing new silicon, software and open systems at its Advancing Ai 2025 event.

Held in San Jose, California, the event showcased how AMD and its partners are building an open Ai ecosystem using the newly launched AMD Instinct MI350 Series accelerators and ROCm 7 software. The company also previewed its next-generation rack-scale hardware roadmap targeting hyperscale and enterprise deployments.

Broader Platform Vision

AMD introduced the Instinct MI350X and MI355X GPUs, which it says offer a fourfold increase in Ai compute performance over the previous generation. The MI350 Series is built for large-scale generative Ai and high-performance computing, with AMD citing a 35-times increase in inferencing capability.

The company also unveiled its next generation Ai rack, codenamed Helios, which will feature Instinct MI400 Series GPUs, Zen 6-based EPYC “Venice” CPUs and Pensando “Vulcano” NICs.

“These announcements mark the next phase of our commitment to open, high-performance Ai infrastructure,” said AMD Executive Vice President Forrest Norrod. “We are enabling customers to scale from single-node experimentation to massive multi-rack deployments.”

The Instinct MI350 Series exceeded AMD’s earlier target of improving energy efficiency by 30 times over five years. The company now aims to deliver a 20-times increase in rack-scale efficiency by 2030.

Software And Developer Focus

AMD highlighted ROCm 7, the latest release of its open-source Ai software stack. The update includes new tools, expanded support for frameworks and APIs and improved developer workflows. It is designed to better support generative Ai and HPC workloads.

Also launched was the AMD Developer Cloud. This managed environment offers developers access to AMD hardware and ROCm software for rapid Ai development and testing.

“We’re removing barriers to Ai compute,” said Brad McCredie, AMD Corporate Vice President. “With ROCm 7 and our developer cloud, teams can begin building at scale using open platforms.”

Industry Adoption

Several partners outlined how they are using AMD’s platforms.

Meta said its data centres already use Instinct MI300X GPUs for Llama 3 and 4 inference, and it plans to integrate MI350 and MI400 in future systems.

Oracle Cloud Infrastructure is adopting AMD’s open rack-scale design with plans to scale up to over 131,000 MI355X GPUs for zettascale Ai clusters. Microsoft also confirmed that MI300X is now powering multiple models on Azure.

Cohere stated its Command large language models are deployed on MI300X platforms, citing high throughput and data privacy. HUMAIN announced it is working with AMD to build scalable and cost-efficient Ai infrastructure using AMD’s full portfolio.

Red Hat and AMD are collaborating to bring AMD Instinct GPUs to hybrid cloud environments using OpenShift Ai. Meanwhile, Astera Labs and Marvell are advancing UALink, an open interconnect technology to help integrate next-gen infrastructure.

Last Updated on June 13, 2025 by Nick Ross