Saudi Arabia’s push to become a major AI infrastructure hub has moved from ambitious capacity targets to live hardware. HUMAIN has begun delivering production AI compute using AMD Instinct MI355X GPUs, AMD EPYC processors and Cisco networking, with a much larger expansion already planned for 2027.
The deployment gives HUMAIN customers access to GPU-as-a-service infrastructure inside Saudi Arabia, supporting workloads including AI model training and inference. AMD’s accelerators are connected through Cisco’s N9000 Series platform, using Silicon One networking technology and 800G optics to handle the high-bandwidth communication required by large GPU clusters.
Getting the first systems into production matters because the bigger numbers surrounding Saudi Arabia’s AI ambitions remain largely forward-looking. AMD, Cisco and HUMAIN now plan to begin deploying as much as 250 MW of additional AI infrastructure in 2027, with capacity expected to start becoming available during the second half of the year.
That next phase is expected to move to AMD’s newer Instinct MI400 Series GPUs alongside EPYC CPUs and the company’s ROCm software platform. Cisco will provide networking and other supporting infrastructure. The companies say their previously announced joint venture remains on course for as much as 1 GW of AI infrastructure by 2030, although reaching that figure will depend on customer demand, construction, hardware availability and the practical challenges of operating data centres at that scale.
The project also reflects a wider shift in the AI market. Countries increasingly view access to compute, data residency and AI models as strategic infrastructure rather than simply cloud services purchased from overseas providers. Saudi Arabia has been particularly aggressive in pursuing that opportunity, using HUMAIN, which is backed by the Public Investment Fund, as a vehicle for building domestic AI capacity.
Sovereign AI is consequently an important part of the pitch. Locally operated infrastructure can give governments and businesses greater control over where sensitive data is stored and how AI systems are deployed. HUMAIN also intends the platform to support open models and software, potentially giving customers more flexibility than tightly integrated proprietary stacks.
AMD has its own strategic incentive. Nvidia remains the dominant supplier of accelerators for large-scale AI computing, and deployments such as HUMAIN provide AMD with an opportunity to establish Instinct hardware and ROCm as credible alternatives for massive AI clusters. Cisco, meanwhile, is competing for the networking layer that connects increasingly large numbers of accelerators.
The immediate milestone is considerably smaller than the infrastructure promised for the end of the decade, but it turns part of that roadmap into something customers can actually use. The more significant test begins in 2027, when the partners will have to demonstrate that hundreds of megawatts of planned capacity can become operational infrastructure rather than remain an impressive headline figure.
