AMD Unveils New AI GPUs and EPYC CPUs for Data Centres

AMD has expanded its AI infrastructure portfolio with a series of hardware and software launches aimed squarely at the evolving requirements of data centres. 

At its Advancing AI event, the company introduced sixth generation EPYC processors, the Instinct MI400 GPU family, its ROCm.ai software platform and a new rack-scale architecture designed for large AI deployments.

Opening the event, Dr Lisa Su, CEO of AMD, outlined how changing AI workloads are reshaping infrastructure requirements across the industry.

“Our mission is to push the boundaries of high performance and AI computing,” Lisa said during the keynote. “I often say AI is the most important technology in the last 50 years.”

She emphasised the scale of demand facing infrastructure providers, noting that the industry now processes more than 35 quadrillion AI tokens every month. Around 60% of available AI capacity is dedicated to inference rather than model training, reflecting growing adoption of agentic AI.

“Compute demand is growing at an incredible pace,” Lisa noted. “When you ask the agent to do something, it has dozens of steps… so you need lots of GPUs to do all that reason and a lot of CPUs to orchestrate every step around it.”

New processors for AI workloads

AMD's sixth generation EPYC 9006 Series is designed to address the mix of workloads now found inside modern data centres. The range includes the EPYC 9006 SP7, EPYC 9006 SP8, 9006X SP7 and 9006 LP processors, with configurations intended for enterprise applications, high-performance computing and AI environments.

The company says agentic AI places increasing demands on CPUs, which coordinate workloads, connect models with enterprise applications and manage orchestration around GPU resources.

Dan McNamara, Senior Vice President and General Manager of Compute and Enterprise AI at AMD, described how the new processors are intended to support those changing infrastructure requirements.

“As agentic AI reshapes how enterprise infrastructure is designed and deployed, EPYC is evolving with our customers once again,” Dan said.

“6th Gen AMD EPYC delivers workload-optimised compute for the full spectrum of modern data centre needs; from running more agents and feeding GPUs across rack-scale AI infrastructure to powering the business-critical applications and services organisations rely on every day.”

Alongside the processors, AMD unveiled the Instinct MI400 GPU family for large AI training clusters and HPC deployments. The portfolio includes the MI455X for hyperscale AI factories and the MI430X for sovereign AI and scientific computing.

“The AMD Instinct MI400 Series extends our AI portfolio with purpose-built solutions optimised for the full spectrum of AI and HPC deployments, built on an open software foundation that gives customers the flexibility to innovate at every scale,” says Vamsi Boppana, Senior Vice President of AI at AMD.

Building a software ecosystem

Hardware formed only part of the announcements. AMD also launched ROCm.ai, a software platform intended to simplify AI deployment and optimisation across AMD infrastructure.

The platform combines a new command-line interface with AI-assisted developer tools and Hyperloom, an open-source inference optimisation system. According to AMD, ROCm.ai delivers up to 3.3 times faster inference performance and up to 2.4 times faster training on average for the latest version of ROCm over ROCm 7 on the same hardware.

“AMD is committed to helping developers move faster than ever,” says Anush Elangovan, Corporate Vice President of AI Software at AMD. “ROCm.ai brings agentic AI development to AMD platforms – agents that don’t just answer questions, but profile, debug and drive workloads toward peak performance on AMD hardware.”

AMD also expanded its collaboration with AT&T and Microsoft through OTel 2.0, an open-source AI model developed specifically for telecommunications networks.

Built using Microsoft Foundry and powered by AMD Instinct GPUs, the model is trained using data from the Open Telco AI initiative to help operators automate network management and improve operational resilience.

Developing rack-scale infrastructure

For operators building larger AI clusters, AMD introduced Helios, a rack-scale infrastructure platform integrating EPYC processors, Instinct MI455X GPUs, Pensando networking and ROCm software.

AMD says the platform provides 50% more high-bandwidth memory capacity and up to 30% more tokens per dollar than competing systems. The modular design enables deployments ranging from individual racks to gigawatt-scale AI campuses.

The company also expanded into physical AI with the launch of its Robotics Partner Network, bringing together hardware vendors, software developers, sensor companies and system integrators.

“No single company will build the future of physical AI,” explains Amey Deosthali, Senior Director of Industrial, Robotics and Healthcare at AMD.

“Building robots requires bringing together AI models, middleware, sensors, simulation tools, system integrators and production hardware. Too often, those pieces come from disconnected vendors with little validation between them.

“With the AMD Robotics Partner Network, we are extending the AMD open ecosystem approach into robotics to help reduce integration efforts, shorten development cycles and allow roboticists to move from reference platform to production systems with greater flexibility, portability and confidence.”

AMD also introduced the Ryzen AI Embedded X100 Series system-on-chip for robotics, manufacturing and healthcare applications, alongside Kria AI Solutions, a development platform designed to speed deployment of autonomous systems.

“The Ryzen AI Embedded X100 Series marks an exciting new chapter for physical AI, unlocking opportunities across a broad range of intelligent, real-time embedded applications,” adds Kirk Saban, Corporate Vice President of Product, Software and Solutions at AMD.