AMD and Cisco have teamed up with Saudi technology firm HUMAIN to launch live Instinct MI355X computing infrastructure in the Kingdom. The deployment pairs AMD accelerators with Cisco Silicon One networking to deliver on demand GPU capacity for training and inference workloads. The partnership establishes the opening stage of a roadmap designed to reach 1 gigawatt of total capacity by 2030.
The active production cluster combines AMD Instinct MI355X GPUs with EPYC server processors. Cisco handles the interconnect layer through its N9000 series platform, which relies on Silicon One silicon and 800G optical connections. This setup creates an open network fabric built for low latency communication between processor nodes, giving enterprise clients direct access to bare metal GPU power.
According to the official joint announcement, this setup allows HUMAIN to operate as an independent sovereign cloud provider. Clients can run massive model training runs or live inference queries without routing sensitive internal data through foreign data centers.
The joint venture plans to ramp up hardware deliveries substantially. A major expansion phase scheduled for 2027 will introduce up to 250 megawatts of capacity running on next generation AMD Instinct MI400 series GPUs and open source ROCm software. Strong enterprise demand across the region has kept the group on track toward its ultimate goal of deploying 1 gigawatt before the end of the decade.
AMD chief executive Lisa Su pointed to the hardware stack as a foundation for regional computing growth:
With AMD Instinct GPUs, EPYC CPUs and ROCm open software at the foundation, we are expanding the compute capacity needed to advance AI innovation across the Kingdom and globally.
HUMAIN chief executive Tareq Amin noted that bringing the cluster into active service confirms the ability of the venture to build and manage top tier computing clusters at scale. Cisco head Chuck Robbins echoed the sentiment, stating that the project gives the region the core networking infrastructure needed to move from artificial intelligence planning into actual daily delivery.
A central pillar of the initiative is data sovereignty. Governments, enterprise customers, and academic labs gain access to advanced hardware while retaining full authority over model training, security rules, and data residency. By avoiding closed proprietary hardware ecosystems, the platform lets regional teams adapt modern artificial intelligence models to local languages and distinct cultural contexts.
