AMD says production artificial-intelligence infrastructure powered by Instinct MI355X GPUs, AMD EPYC CPUs and Cisco networking is now live and serving customers in Saudi Arabia. The company also says its partners plan to deploy up to 250 megawatts of additional AI infrastructure from 2027. AMD also says the joint venture remains on track to reach up to 1 gigawatt by 2030.

The claims were published by AMD on September 9, 2026, in an article about its work across the Middle East. They describe a shift from announced investment and planned capacity towards what AMD characterises as an operating, customer-serving system. The company has not disclosed the infrastructure’s exact location, customer list, number of processors, workloads, utilisation or delivered computing capacity.

Contents

What changed in Saudi Arabia

AMD, Cisco and HUMAIN announced that the Saudi production system is live and serving customers. The stated hardware stack combines AMD Instinct MI355X graphics processing units (GPUs), AMD EPYC central processing units (CPUs) and Cisco networking.

A GPU is a processor designed to perform many calculations in parallel. That makes GPUs useful for accelerating AI model training and inference, the process of using a trained model to generate predictions or responses. CPUs are general-purpose processors used for operating-system tasks, application logic and coordination across a server.

The announcement does not establish how much of the planned Saudi capacity is already operational. AMD says partners will begin deploying up to 250 megawatts of additional AI infrastructure in 2027. It also says the joint venture remains on track to reach up to 1 gigawatt by 2030.

Those figures describe power capacity or infrastructure scale, not directly the number of GPUs, the models being served or the amount of useful AI work completed. One gigawatt equals 1,000 megawatts, but the source does not say whether the figures refer to IT load, total facility power or another measure.

How the infrastructure fits together

The Saudi announcement illustrates that large AI systems are not built around a single accelerator chip. A data-centre deployment typically combines AI accelerators, general-purpose processors, networking, storage and software.

In this case, the MI355X GPUs are described as the AI acceleration component, while EPYC CPUs handle general server functions and coordinate workloads. Cisco networking connects the computing systems and supports communication between components. AMD presents this broader combination as part of a portfolio spanning CPUs, GPUs, adaptive and embedded computing, networking and open software.

That integrated approach matters because AI deployment involves more than training a model. Operators must move data between processors and storage, connect servers, manage applications and provide access to users or businesses. The announcement, however, does not provide benchmark results, energy-efficiency measurements, service-level information or a comparison with competing systems.

An enterprise entry point for AI deployment

AMD also describes the HUMAIN AI Application Platform, referred to in the article as “AI-in-a-Box”. The platform combines AMD Instinct GPUs and EPYC CPUs with HUMAIN’s capabilities.

AMD positions it as a starting point for Saudi enterprises that want to test workloads, deploy initial AI tools and expand later without beginning with hyperscale infrastructure.

This is different from the national-scale capacity target. A large infrastructure programme can provide computing resources, while an enterprise platform is intended to make those resources usable by individual organisations with more limited or specific needs. The source does not disclose the platform’s software architecture, pricing, model support, service availability or technical limits.

AMD, Saudi Arabia’s Ministry of Communications and Information Technology, and the Digital Cooperation Organization also announced an open developer ecosystem initiative. AMD says the programme will provide developers, startups and software engineers with open software tools, training, technical expertise and responsible-AI governance resources.

The announcement does not specify the applicable governance rules, standards, access conditions or participating developers. It therefore establishes the proposed scope of the initiative, rather than demonstrating its adoption or impact.

The wider regional programme

AMD says its Middle East activities extend beyond Saudi Arabia:

  • In the United Arab Emirates, the company says it is collaborating with e& across AI-ready cloud, telecommunications, enterprise and edge infrastructure.
  • With Core42, a G42 company, AMD says it is advancing AI, machine learning and confidential computing for cloud environments. Confidential computing uses hardware- and software-based protections intended to help safeguard data while it is being processed; the specific protections depend on the system.
  • In Qatar, AMD says it is working with MEEZA to accelerate AI adoption and exploring the use of AMD technologies in MEEZA data centres.

These statements describe collaborations and areas of exploration. The supplied announcement does not identify concrete production deployments, delivery milestones or operating customers for the UAE and Qatar work.

Together, the partnerships point to a regional deployment model involving governments, data-centre operators, cloud providers, telecommunications companies, networking suppliers and application platforms. They also reflect the idea of sovereign AI: developing or operating AI infrastructure, data and capabilities under a country’s or organisation’s control and according to its governance and security requirements.

What happens next

The next major milestone is the planned deployment of up to 250 megawatts of additional AI infrastructure from 2027. Beyond that, AMD says the joint venture remains on track for an eventual target of up to 1 gigawatt by 2030.

As the programme expands, further details about the joint venture, customer access and the AI-in-a-Box and developer initiatives will help show how the infrastructure is being used. The scale of the regional programme will become clearer as additional capacity comes online and the participating organisations publish deployment details.

The announced direction is therefore a phased build-out: a production system already serving customers in Saudi Arabia, an additional deployment planned from 2027, and a longer-term regional capacity target for 2030.

Sources