Humain signs Nebius for up to 50 MW as it grows its Nvidia chip commitment

PIF-owned Humain used Nvidia’s GTC conference in Berlin to show the demand side of its AI buildout, naming its first international anchor customer and expanding its chip order from Nvidia.

AI cloud provider Nebius has signed a preliminary framework to take up to 50 MW — half the initial 100 MW now under construction — at a planned AI giga-campus in Neom’s Oxagon being developed by Humain and DataVolt, the digital-infrastructure firm that will own and operate the site, according to a statement. Capacity would phase in from 4Q 2027, the companies said in a statement. The framework comes six weeks after ground broke on the site, whose first phase is designed for 360 MW.

Humain also expanded its Nvidia collaboration to as much as 84 MW of compute, some 27k GPUs. It begins with 14 MW of Blackwell HGX B300 systems (c. 6k GPUs) in 4Q 2026 for Humain Cloud, followed by 70 MW of next-generation Vera Rubin infrastructure (c. 21k GPUs) from mid-2027 — including a 50 MW block set aside for an unnamed US client.

REMEMBER- The company is also reworking how it pays for all this. Humain is now targeting 1.7 GW of capacity in 2027, above its earlier plans, and CEO Tareq Amin said it wants to reduce its reliance on shareholder funding. It is pursuing a domestic data-center fund seeking an initial USD 2.5 bn, subject to regulatory approval.

Why it matters: A program of Humain’s size — 3 GW of data-center capacity targeted by 2030 and more than 6 GW by 2034 — needs paying tenants to justify the spend. Humain has said it intends to lock in customer demand before building out, then use that contracted demand to attract financing and outside investment. Landing a foreign anchor this early, even on a preliminary framework, starts to turn announced megawatts into financeable capacity and puts the first international demand behind Saudi Arabia’s pitch to export compute on the back of cheap power.