AMD, Cisco, and HUMAIN said on August 31, 2026 that AMD Instinct MI355X GPU infrastructure is now live in Saudi Arabia and serving HUMAIN customers in the Kingdom and beyond. The production cluster runs on AMD EPYC CPUs and a Cisco Silicon One fabric, using the Cisco N9000 series and 800G optics to tie the GPUs together. HUMAIN is selling the capacity as GPU-as-a-service for training and inference.

The three companies already had a joint venture on the books. This release is the first time they say the MI355X hall is in production rather than planned. They also restated the next phase: up to 250 megawatts of infrastructure based on AMD Instinct MI400 Series GPUs, EPYC CPUs, AMD ROCm software, and Cisco networking, with deployment starting in 2027 and capacity expected to come online in the second half of that year. The longer target remains up to 1 gigawatt in Saudi Arabia by 2030.

“Bringing AMD Instinct systems online in Saudi Arabia is an important milestone in our work with HUMAIN.” Dr. Lisa Su, chair and CEO, AMD

“Demand for sovereign, high-performance AI infrastructure continues to grow, and together with AMD and Cisco, we are expanding Saudi Arabia’s role as a platform serving customers across the Kingdom, the region and globally.” Tareq Amin, CEO, HUMAIN

“The region is moving quickly from AI investment to delivery, and through our work with HUMAIN and AMD, Cisco is building the critical infrastructure to make AI vision a reality, in the Kingdom and beyond.” Chuck Robbins, chair and CEO, Cisco

Open stack, local control

The companies said the platform is meant for governments, enterprises, research labs, and developers that want open models and open software on locally operated machines. AMD supplies the compute and ROCm. Cisco supplies an interoperable network with centralized control as the cluster grows. HUMAIN operates the site and the customer relationship. The pitch is that a buyer can keep data in-country, customize models, and still rent modern GPUs without standing up a private factory.

Cisco published the same announcement from its own newsroom. Neither release disclosed how many MI355X GPUs are installed, which customers are already on the service, or the megawatts in the live hall.

Decoded Take

A live MI355X cluster is a different claim from another Riyadh memorandum. It puts AMD in a sovereign GPU cloud that is usually narrated as an NVIDIA story, and it gives Cisco a production reference for Silicon One in an AI factory. The 250 megawatt MI400 phase and the 1 gigawatt 2030 line are still future slides. The test is whether HUMAIN can name paying tenants, including buyers outside the Kingdom, before those later halls exist. If the customer list stays quiet, this is a commissioned island. If it fills, Saudi Arabia has a second-source GPU cloud that is actually running.