Cisco and G42 to deploy AMD MI350X GPUs for new AI cluster
Cisco will power, connect, integrate, and secure a large-scale AI cluster for G42
#UAE #datacentres - Abu Dhabi’s AI powerhouse G42 and US networking and security giant Cisco have announced a major infrastructure collaboration to deploy a large-scale AI cluster featuring AMD Instinct MI350X GPUs within the planned 5 Gigawatt (GW) UAE-US AI Campus. Cisco will provide end-to-end infrastructure including compute, networking, security, storage and optics, whilst also serving as technology integrator within G42’s Regulated Technology Environment compliance framework. The deployment builds on the US-UAE AI Acceleration Partnership, whose flagship projects include the 1GW Stargate UAE cluster (currently under construction in Abu Dhabi) and the 5GW campus announced during President Trump’s May 2025 visit to the UAE. No details about the new clusters’ capacity or deployment schedule were shared by Cisco or G42.
SO WHAT? - Although the capacity is so far not been made public, the planned deployment of MD Instinct Series GPUs by G42 could be one of the biggest announced so far. In June, AMD and Saudi Arabia’s HUMAIN announced that their global data centre joint venture would deploy MI350s. Meanwhile, earlier this month OpenAI signed a 6GW agreement with AMD to use Instinct GPUs, beginning with MI450s in the second half of 2026. AMD’s new Instinct GPUs are positioned as direct competitors to NVIDIA H100 and H200, offering total cost of ownership (TCO) advantages. AI infrastructure vendors involved in the UAE-US AI Campus include AMD, Cerebras and NVIDIA.
Here are some key points about the announcement:
G42 and Cisco have announced a major infrastructure collaboration to deploy a large-scale AI cluster with AMD Instinct MI350X GPUs. The new cluster will form part of the planned 5 Gigawatt (GW) UAE-US AI Campus and the networking giant will provide end-to-end infrastructure including compute, networking, security, storage and optics. Cisco will also serve as technology integrator within G42’s Regulated Technology Environment compliance framework.
The deployment integrates Cisco’s full-stack AI infrastructure including UCS 885A servers with AMD MI350X GPUs for compute, Nexus 9K 800G switches for networking, Firepower 4200 firewalls for security, and advanced optics alongside observability and analytics tools.
AMD Instinct MI350X GPUs are built on CDNA 4 architecture and provide 288GB HBM3E memory from Micron and Samsung Electronics with up to 8TB/s bandwidth, supporting flexible air-cooled configurations of up to 64 GPUs or direct liquid-cooled racks with up to 128 GPUs.
The MI350Xs deliver up to four times generation-on-generation AI compute improvement and up to 35 times enhancement in inferencing performance.
Cisco will act as technology integrator within G42’s Regulated Technology Environment, a compliance and security framework ensuring advanced compute infrastructure operates with protection, transparency and governance to prevent unauthorised access, transfer or misuse.
The collaboration advances the US-UAE AI Acceleration Partnership, with flagship projects including the 1GW Stargate UAE cluster under construction in Abu Dhabi and the 5GW UAE-US AI technology campus announced in May 2025.
G42 and Cisco will coordinate with US government agencies to obtain necessary licences and regulatory clearances for deployment, which is expected to advance in coming months pending final approvals.
AMD launched the Instinct MI350 Series in June 2025, with the GPUs delivering up to 1.3 exaFLOPS of MXFP4/MXFP6 performance and availability through cloud service providers including major hyperscalers and on-premises platforms from Dell, HPE and Supermicro.
ZOOM OUT - The planned Cisco-G42 large-scale AI cluster deployment using AMD Instinct GPUs forms part of the ambitious 5GW UAE-US AI Campus announced during President Trump’s state visit to Abu Dhabi in May 2025. Spanning 10 square miles, the campus is designed to be the largest AI datacentre deployment outside the United States and will deliver sovereign AI-grade compute at scale to serve nearly half the world’s population. G42 is leading construction and operation in partnership with a number of US companies, utilising nuclear, solar and gas power to minimise environmental impact. The facility will enable US hyperscalers to provide low-latency services across the surrounding region.
[Written and edited with the assistance of AI]
Read more about the planned UAE-USA AI Campus in Abu Dhabi:
Cerebras to expand AI infrastructure to the UAE (Middle East AI News)
US approves billions in Nvidia UAE sales (Middle East AI News)
Tech leaders unite to build Stargate UAE AI hub (Middle East AI News)
New US-UAE agreement sure to accelerate G42’s US plans (Middle East AI News)
UAE-US to launch AI cluster with 5GW data centre (Middle East AI News)
This is a signifcant deployment for AMD! The MI350X series represents a real competitive alternative to NVIDIA for hyperscale deployments. The collaboration between Cisco, G42, and AMD within the UAE-US AI Campus framework shows how quickly the AI infrastructure landscape is diversifying. The 288GB HBM3E memory and 8TB/s bandwidth specs are exactly what's needed for large model training. Very interested to see how the TCO advantages play out at the 5GW scale!