Saudi Arabia’s Humain is not thinking small. The Public Investment Fund-owned AI company, launched in May 2025 under Crown Prince Mohammed bin Salman, is reportedly working to establish a fund of up to $3 billion dedicated to expanding its data center footprint, adding to a financing push that already spans Goldman Sachs, the National Infrastructure Fund, and a growing list of strategic partners.
The scale is genuinely hard to overstate
Humain’s stated goal is 6 GW of AI data center capacity by 2034, with an interim target of 1.9 GW by 2030. For context, a single gigawatt can power roughly 750,000 average US homes. Humain is planning to dedicate multiples of that to training AI models.
The company has already secured land access for up to 14 GW of power across 211 plots, suggesting the infrastructure groundwork is well underway rather than purely aspirational.
On the financing side, the company locked in a $1.2B framework with Saudi Arabia’s National Infrastructure Fund in January 2026, earmarked for 250 MW of capacity. Goldman Sachs is advising on a separate financing package potentially exceeding $5.33B, covering both data centers and GPU procurement.
Partnerships, contracts, and a NEOM campus
In August 2026, the company awarded contracts to Al Moammar Information Systems to add 200 MW of new data center capacity, effectively quadrupling the existing footprint under that partnership.
Separately, a collaboration with DataVolt is taking shape at NEOM’s Oxagon district, where a planned 1.5 GW campus is under development. Humain is contributing 100 MW to that project.
On the investment side, Humain committed $3B to Elon Musk’s xAI and struck a deal with Blackstone-backed AirTrunk on a $3B data center campus.
Humain is also developing its own large language models, including one called Allam, built around Arabic-language capabilities. The logic is straightforward: global AI infrastructure skews heavily toward English, and there are roughly 400 million Arabic speakers whose linguistic and cultural context remains underserved by existing frontier models.