AMD, Cisco, and the AI ecosystem Humain have announced the launch of an AI infrastructure based on AMD Instinct MI355X accelerators, AMD EPYC processors, and Cisco networking equipment. This system is already in use by Humain’s clients and supports its GPU-as-a-Service offering, from model training to inference.
The MI355X accelerators are connected through the Cisco N9000 Series platform using Cisco Silicon One networking technology and 800G optical connections. This configuration is designed for large cluster operations, where performance relies not only on the computational power of GPUs but also on the data exchange speed between accelerators.
Image source: generated by Nano Banana
The next phase of the joint project involves the construction of up to 250 MW of additional AI infrastructure starting in 2027. It is planned to utilize AMD Instinct MI400 Series accelerators, EPYC processors, and the ROCm software stack along with Cisco’s networking solutions. Initial capacities from this phase are expected to begin coming online in the second half of 2027.
Meanwhile, AMD, Cisco, and Humain maintain a long-term goal of increasing the combined capacity of their joint venture in Saudi Arabia to 1 GW by 2030. The 250 MW and 1 GW figures are stated growth plans, not the volume of infrastructure already installed; the currently operational facility features the MI355X system.
The companies did not disclose the size of the operational cluster, its load, clients, the cost of GPU-as-a-Service, or service terms.