At the AI 2025 event, AMD officially announced its artificial intelligence strategy and new infrastructure. The company aims to provide comprehensive solutions in this field with an integrated product range—from network solutions to open software platforms.

AMD showcases its new AI technologies

The new-generation GPU series introduced by AMD, named Instinct MI350, delivers four times the AI compute power compared to the previous generation. Comprising the MI350X and MI355X models, the series also achieves a 35-fold performance improvement in inference tasks.

It was highlighted that the MI355X model offers 40% better “dollar-per-token” efficiency compared to competing solutions. These advancements provide significant gains in terms of cost and scalability in AI applications.

AMD also introduced a new open-standard-based AI infrastructure centered around these GPUs. This infrastructure includes 5th Gen EPYC processors and Pensando Pollara NIC cards. These solutions are expected to see broader deployment in the second half of 2025.

At the same event, AMD also introduced its next-generation rack system called Helios. The Helios system will run on “Venice” EPYC processors based on the Zen 6 architecture, Pensando “Vulcano” network cards, and the next-generation Instinct MI400 GPUs. It was announced that this new system could offer up to a 10x performance boost in inference tasks for Mixture of Experts (MoE) models compared to the previous generation.

On the software side, AMD announced the release of ROCm 7, its new open-source software stack designed to meet the demands of generative AI and high-performance computing. ROCm 7 improves compatibility with industry-standard frameworks and expands hardware support.

Impressive data regarding energy efficiency was also shared. AMD has surpassed its original 2020 goal of achieving a 30x improvement in energy efficiency for AI training and high-performance computing nodes—reaching a 38x improvement.