HUMAIN brings AI infrastructure online with AMD, Microsoft and Together AI deals at LEAP 2026
AMD Instinct GPUs on Cisco networking went into production for HUMAIN customers, part of a plan for 1GW of Saudi AI capacity by 2030, alongside separate Microsoft and Together AI deals.
- Compute & infrastructure
- Money & business
- Notable
At LEAP 2026, the technology conference in Riyadh, Saudi Arabia’s state-owned HUMAIN announced three separate infrastructure deals with AMD, Microsoft and Together AI, moving from announced partnerships toward hardware actually in production.
AMD and Cisco said AMD Instinct MI355X GPUs, paired with EPYC CPUs and Cisco Silicon One networking, had gone live and were “serving HUMAIN customers in the Kingdom and beyond,” describing it as the first phase of a plan reaching up to 1 gigawatt of AI infrastructure capacity in Saudi Arabia by 2030. A second phase, using AMD’s next-generation Instinct MI400 series, is planned to bring up to 250MW of additional capacity online from 2027.
Microsoft and HUMAIN announced HUMAIN ONE, an agentic AI layer for enterprise workflows, to be bundled with Microsoft 365 Copilot on Azure and targeted initially at one million users across the Middle East and Africa. The companies also said a Windows-based HUMAIN AI PC would go on sale to enterprise customers on 20 September, with a goal of one million such devices sold across the region by 2030, and that HUMAIN’s Arabic-language ALLAM models would become available through Microsoft Foundry.
Separately, Together AI and HUMAIN announced a 250MW datacentre in Saudi Arabia intended to connect HUMAIN’s compute capacity to global developer demand for inference and training, which Together AI projected would generate more than $5 billion in gross annualised revenue in its first year — a projection from the companies themselves, not yet a reported result.
The cluster of announcements followed Mistral’s own sovereign-AI partnership with HUMAIN the previous week, underscoring HUMAIN’s strategy of signing parallel deals with multiple Western chipmakers, cloud providers and model developers simultaneously rather than committing to a single infrastructure partner, while positioning Saudi Arabia’s power capacity and geography as unused capacity for global AI compute demand.