#SaudiArabia #LEAP5 - AMD, Cisco and Saudi Arabia’s PIF-owned AI company HUMAIN have announced that AMD Instinct MI355X GPU-based AI infrastructure, built on Cisco Silicon One networking, is now live in production in Saudi Arabia and serving HUMAIN customers in the Kingdom and beyond. According to Lisa Su, CEO AMD, the site is AMD’s largest inference cluster outside the USA. The three technology companies plan to deploy up to 250 megawatts of further AI infrastructure starting in 2027, part of a joint venture on track to reach 1 gigawatt of AI infrastructure by 2030.
SO WHAT? - Announcements about planned AI infrastructure are common; this one confirms the first phase of actual production deployment. GPU-as-a-service is now running live for real customers on Saudi soil and the companies say demand is what’s driving the next 250MW phase. That’s a stronger signal of genuine enterprise appetite for sovereign AI compute in the region than the underlying capacity figures alone.
KEY POINTS:
AMD, Cisco and HUMAIN have announced that new AMD Instinct MI355X GPU-based AI infrastructure, is now live in Saudi Arabia. According to Lisa Su, CEO AMD, it is the largest AMD inference cluster outside of the USA. HUMAIN CEO Tareq Amin says that 13,000 AMD GPUs will be deployed this year.
The live deployment combines AMD Instinct MI355X GPUs and AMD EPYC CPUs with Cisco Silicon One-based networking and 800G optics, interconnected through the Cisco N9000 Series platform to support model training and inferencing at scale.
The next phase of the joint venture will deploy up to 250 megawatts of AI infrastructure powered by AMD’s newer Instinct MI400 Series GPUs and ROCm open software, alongside Cisco networking, with capacity expected to begin coming online in the second half of 2027.
Dr Lisa Su, chair and CEO of AMD, said the milestone represents an open, high-performance AI platform already serving customers today, with compute capacity set to expand significantly in the coming years.
Tareq Amin, CEO of HUMAIN, said the production deployment demonstrates HUMAIN’s ability to build, operate and deliver advanced AI infrastructure at scale, driven by growing demand for sovereign, high-performance AI infrastructure.
Chuck Robbins, Chair and CEO of Cisco, said the region is moving quickly from AI investment to delivery, with Cisco providing the critical networking infrastructure to support that shift.
The platform is designed to give governments, enterprises, research institutions and developers control over where their data resides, how models are customised, and how AI systems are deployed and governed, built on open models and open software rather than closed proprietary systems.
The collaboration builds on a previously announced joint venture between AMD, Cisco and HUMAIN, which remains on track to deploy up to 1 gigawatt of AI infrastructure in Saudi Arabia by 2030.
HUMAIN AI-in-a-box - AMD and HUMAIN also announced ‘AI in a Box’ on the first day of LEAP 2026. a turnkey AI compute solution aimed at Saudi Arabian enterprises just starting their AI adoption journey. Built by HUMAIN using AMD Instinct GPUs and EPYC CPUs, the package targets sectors including oil and gas, financial services and manufacturing, letting companies begin deploying AI and testing use cases without committing to hyperscale infrastructure from the outset. The launch builds on AMD and HUMAIN's broader collaboration to invest up to $10 billion in Saudi AI infrastructure, part of the joint venture with Cisco targeting up to 1 Gigawatt of AI capacity in the Kingdom by 2030.
ZOOM OUT - The GPUs now running at HUMAIN's facility are AMD's current flagship AI accelerator. Built on AMD's 4th Gen CDNA architecture, the Instinct MI355X packs eight GPUs onto a single baseboard, each with 288GB of HBM3E memory and 8TB/s of bandwidth, combining into 2.3TB of shared memory across the platform. This allows users to train and run very large AI models without constantly shuffling data in and out of memory. Direct liquid cooling lets each GPU draw up to 1,400W, pushing higher clock speeds than air cooling alone would allow, while newer MXFP6 and MXFP4 datatype support improves inference speed and power efficiency for generative AI workloads specifically. The chips already power some of the largest AI deployments in the world, including Meta's Llama 405B and GPT models.
[Written and edited with the assistance of AI]
Source: AMD, Cisco, HUMAIN
Read more about the AMD-HUMAIN partnership:
Luma AI secures $900M in Series C round led by HUMAIN (Middle East AI News)
HUMAIN & AMD target 6GW of AI compute by 2034 (Middle East AI News)
Saudi forms AI-focused $10B joint venture with AMD (Middle East AI News)