Humaine’s investments put infrastructure on the line, $15 billion to link computing models with energy networks

Riyadh opened the LEAP 2026 conference at the Malham Exhibition and Conference Center under the slogan “Towards New Worlds”, announcing on its first day deals and technology launches worth more than $15 billion, a step that underscores the kingdom’s aim to own the AI infrastructure joints rather than remain in consumer seats. Humaine, owned by the Public Investment Fund, leads these moves as the sovereign arm headed by Tarek Amin, former CEO of Aramco Digital, to deliver computing, energy and data-center commitments locally.

The most prominent element of these announcements is a $10 billion partnership linking Humaine with AMD and Cisco, in which Instinct MI355X graphics processors and EPYC CPUs have entered actual operation in Saudi Arabia alongside Cisco Silicon One networks to serve the company’s customers. The partnership plans to deploy up to 250 megawatts of computing infrastructure starting in 2027, on a path that aims to reach a full gigawatt by 2030, moving the commitments from conference announcements to real hardware on the ground.

In parallel, Amazon Web Services set the launch date for its first cloud region in the kingdom for December 2026, with investments exceeding $5.3 billion, and expanded its cooperation with Humaine to create the first AI region delivering up to 50 megawatts by 2028, based on Trainium chips and Nvidia’s infrastructure built for training and inference workloads. XAI also outlined plans to roll out a data centre starting at 50 megawatts and scaling toward 500 megawatts in the first external deployment of Grok models outside the United States.

The real decisive factor in digital sovereignty is not the announcement of models, but the ability to supply data centres with power and secure actual electrical connectivity.In this context, the Saudi Electricity Company signed an agreement to feed Humaine’s data-center project in the Saad area of Riyadh through the National Grid Company, an operational step that makes computing a material reality beyond silent racks and devices.

On the model side, Humaine placed its M3 model in a limited preview on the Node platform, a mixture-of-experts system containing 428 billion parameters with 23 billion activated per token, trained on more than a trillion Arabic tokens, and promised to release its weights openly later. It also introduced a Voice system to handle dialect families in Saudi Arabia, Egypt, the Maghreb and the Levant, an attempt to close the performance gap that global systems face with Arabic linguistic diversity.

This shift directly impacts business realities in the Gulf, Egypt and the Levant, as providing local computing and inference capacities of hundreds of megawatts re-prices model operating costs for enterprises, banks and regional software firms, and reduces reliance on cross-border external clouds. The situation forces local tech teams to acquire hybrid-hardware management skills and adapt inference frameworks to run concurrently on AMD and Trainium chips instead of relying on a single-technology stack, while also giving developers an internal cloud architecture that meets localisation requirements and enables services built on genuine dialectal understanding.

The most important test after the exhibition lights go out remains the time gap between announced schedules and final delivery.The projects span a timeline from 2026 to 2030, and success is measured by customers’ ongoing willingness to pay for genuine computing and operational services that prove their long-term viability.