xAI's first international megacluster is heading to Saudi Arabia

Elon Musk's xAI is packing its bags for its first international supercluster.

At the U.S.–Saudi Investment Forum in Washington, D.C., xAI inked a major framework deal with HUMAIN, an AI entity owned by Saudi Arabia’s Public Investment Fund (PIF). The plan is simple but massive: build a 500 MW+ flagship hyperscale data center in the Kingdom and roll out Grok across the entire country.

This marks xAI's first large-scale compute footprint outside the United States, expanding far beyond its current superclusters in Memphis.

18,000 GB300s in the desert

The centerpiece of the agreement is an ultra-dense compute site designed to house approximately 18,000 NVIDIA GB300 GPUs in its initial cluster. These chips will be dedicated to training and inferencing future generations of Grok.

Beyond the physical data center, HUMAIN and xAI are co-building a national compute fabric connecting multiple AI sites across Saudi Arabia.

The deployment goes all the way up the software stack. Grok will be rolled out nationwide as a foundational AI layer for government entities and private businesses.

It will also be baked directly into HUMAIN ONE, HUMAIN's autonomous agent platform, aiming to power workplace automation across the region.

Saudi Arabia's neocloud bet

This deal isn't happening in a vacuum. HUMAIN has already lined up infrastructure agreements this year with AMD, NVIDIA, and Qualcomm.

Saudi Arabia is executing a very deliberate playbook: combine cheap energy, endless desert land, and sovereign wealth capital to turn the country into a global exporter of raw AI compute.

For xAI, which has been consuming electricity at record pace to keep up with OpenAI and Google, finding a partner willing to hand over 500 megawatts without utility gridlocks is pure gold.

Why it matters

AI compute has officially outgrown the local constraints of traditional tech hubs. By securing 500 MW+ of capacity and 18,000 GB300 GPUs, xAI is ensuring its next frontier training run won't stall out over domestic power shortages.

For users and developers, this means faster Grok release cycles and massive inference bandwidth. If you operate in the MENA region, Grok's native integration into HUMAIN ONE will likely make it the default enterprise model for regional workflows.

We think this partnership is a masterclass in pragmatic scaling, but it doesn't come without friction. Operating top-tier NVIDIA silicon in the Gulf requires navigating tight U.S. export controls and geopolitical scrutiny. But as the race for compute hits physical energy limits, going where the power flows freely looks like a necessary gamble.