HUMAIN, a Public Investment Fund company, has launched production AI infrastructure in Saudi Arabia built on AMD Instinct MI355X GPUs, AMD EPYC CPUs, and Cisco networking. Announced on August 31, 2026, alongside AMD and Cisco, the operational deployment offers GPU-as-a-service for model training and inference workloads within the Kingdom and abroad.
Live fabric and locally hosted software
The operational cluster links AMD Instinct MI355X accelerators and AMD EPYC host processors across Cisco N9000 Series switches. Cisco pairs Silicon One switching silicon with Cisco 800G optics to create an AI-optimized fabric designed for scale, low latency, and operational resilience.
Software integration relies on AMD's ROCm open stack, intended to let enterprise, research, and government developers run open models while retaining data residency, model customization, and governance controls within locally hosted environments. The joint announcement does not specify the active GPU count or capacity of the initial deployment.
Planned buildout phases
The companies plan to expand infrastructure through their joint venture across multi-year phases:
| Phase | Compute and network stack | Disclosed capacity | Timeline |
|---|---|---|---|
| Initial deployment | AMD Instinct MI355X, EPYC CPUs, Cisco Silicon One, N9000 switches, 800G optics | Unspecified | Live (August 2026) |
| Next expansion | AMD Instinct MI400 Series GPUs, EPYC CPUs, AMD ROCm, Cisco networking | Up to 250 MW | Deployment starts 2027; first capacity expected H2 2027 |
| Long-term roadmap | AMD, Cisco, and HUMAIN joint venture infrastructure | Up to 1 GW | Targeted by 2030 |
The next phase introduces AMD Instinct MI400 Series hardware, with initial capacity scheduled to come online in the second half of 2027. Reaching the long-term 1 GW target by 2030 remains subject to ongoing deployment schedules, facility execution, and vendor delivery.