AMD Unveils Helios Platform, Instinct MI400 GPUs, and EPYC CPUs for the Age of Inference and Agents

AMD has presented the portfolio of solutions with which it will respond to the demand for artificial intelligence (AI), which includes the Helios open rack-scale AI platform, AMD Instinct MI400 Series GPUs and the sixth generation of EPYC CPUs.

The use of artificial intelligence is only growing, as AMD’s executive director, Lisa Su, made clear at the ‘Advancing AI’ event, where she highlighted that in the last two years the use of monthly tokens has increased 158 times, and today we see more than 35,000 billion tokens consumed each year.

In this context, Su has also highlighted the demand for computing, with agents as the next big step for artificial intelligence. This requires a greater number of GPUs, while the accelerator market is also growing and is expected to reach $1.4 billion by 2030; but also CPUs for servers, with an estimated growth for 2030 of more than 200 billion dollars.

To respond to the needs of artificial intelligence, the semiconductor firm has presented a wide range of comprehensive AI and high-performance computing solutions that range from silicon to software, inference and agents.

One of them is the Helios open rack-scale AI platform, designed to support large-scale inference, advanced model training and fine-tuning, delivering exaflop-class AI performance.

The unified architecture integrates 5th Generation AMD Instinct GPUs, 6th Generation AMD EPYC Server CPUs, AMD Pensando Networking, and AMD ROCm Software. With this, it provides in a single rack up to 2.9 peak exaflops in FP4, 1.4 peak exaflops in FP8, 31 terabytes of HBM4 memory and 1.7 petabytes / second of memory bandwidth.

Each rack integrates 18 four-GPU compute trays aligned widthwise, supporting 72 GPUs in total. At cluster scale, AMD Pensando Vulcano 800 AI NICs deliver the high-bandwidth, low-latency connectivity needed to support distributed inference, edge model training, and rapid hyperscale growth.

This announcement is accompanied by the launch of AMD Instinct MI400 Series GPUs and 6th Generation EPYC CPUs, both components aimed at data centers, artificial intelligence factories and high-performance computing.

Instinct MI400 Series combines purpose-built hardware, open source software and scalable system design. The entire range offers high memory bandwidth to support the largest AI models and improve efficiency at scale, highlighting the Instinct MI430X model, which offers up to 288 TFLOPS of FP64 performance.

When it comes to EPYC CPUs, the sixth generation is aimed at high-performance AI workloads and focuses on three specific tasks: running agents, powering accelerators, and powering the business services used by agents.

And for artificial intelligence development on AMD platforms, the company has announced ROCm.ai, a native development platform that combines AI-assisted development, intelligent deployment and software optimization to quickly go from idea to production. It can be integrated into leading AI coding assistants such as Claude, Cursor, and Codex.