Humain expands Nvidia deal to 84MW of AI computing capacity now

Humain has expanded its Nvidia partnership to 84MW of AI computing capacity. The PIF-owned Saudi company plans to deploy about 27,000 Nvidia GPUs across two phases. The first is 14MW of Blackwell HGX B300 systems, roughly 6,000 GPUs, with deployment from the fourth quarter of 2026. The second is 70MW of Vera Rubin systems from mid-2027, of which 50MW is for an unnamed US customer. That phase still requires final agreements. The announcement was made at Nvidia’s GTC Berlin event. The 14MW Blackwell capacity will support Humain Cloud, serving AI companies, public entities and local organizations. It will use Quantum-X800 InfiniBand networking. The Vera Rubin phase is split: 50MW dedicated to frontier-scale AI workloads for the US client, targeted to begin in June 2027, and 20MW for Humain Cloud, targeted for July or August 2027. Both Vera Rubin deployments remain subject to final agreements and deployment plans. No contract value was disclosed. This expansion builds on a partnership announced in May 2025 during a US presidential visit to Riyadh. That earlier plan included an initial 18,000 Blackwell processors and a broader goal of up to 500MW of AI computing capacity over five years. Humain has also stated a longer-term target of nearly 3GW of data-center capacity by 2030 and more than 6GW by 2034. The 84MW addition is a step inside those larger figures. It is not the full target. For a Saudi or regional AI user the useful distinction is between capacity that is already committed and capacity that is still subject to final agreements. The 14MW Blackwell phase has a near-term deployment window. The 70MW Vera Rubin phase, including the 50MW US portion, does not yet have final agreements. A reader can see the planned megawatts. A reader cannot yet see the signed contracts or the commissioning dates for the larger share. The presence of a US customer for 50MW means part of the capacity is for export rather than solely for domestic or regional use. Whether that customer is a hyper scaler, a model lab or another buyer is not stated. The name is not given. The allocation is given. A further limit is power and cooling. 84MW is a substantial continuous load. Saudi Arabia has announced large data-center and energy plans, but the specific power source, cooling method and grid connection for this 84MW are not detailed in the announcement. Those details determine whether the planned GPUs can run at the stated scale. Tareq Amin is chief executive of Humain. The company is owned by the Public Investment Fund. The expansion adds a concrete GPU and megawatt figure to its Nvidia relationship. It does not yet add a delivered and operational record for the Vera Rubin portion. Humain opened its first AI cloud on Nvidia HGX B300 systems in September 2026, with the Ministry of Commerce among early users. The new 14MW phase extends that existing Blackwell footprint. The Vera Rubin systems are a later generation. Their arrival depends on the final agreements. Other Gulf markets are also building large AI capacity. The UAE has announced multi-gigawatt projects. Humain’s 84MW is smaller than those longer-term targets but is tied to a specific Nvidia roadmap and a named timeline for the first phase. The comparison is between announced capacity and capacity that has a near-term delivery window. Partners including Dell Technologies, HPE, Lenovo, Schneider Electric and Vertiv are expected to support delivery of the systems. Their role is logistical and engineering. It does not change the fact that the larger Vera Rubin portion still awaits final agreements. The 50MW US allocation is the largest single customer commitment described in the expansion. It sits inside the portion that still needs final agreements. Until those agreements are signed, the 50MW remains a planned allocation rather than a confirmed offtake. The 20MW for Humain Cloud is in the same position. Both depend on the same unresolved step. If Humain publishes the date the 14MW Blackwell systems are live and the status of the final agreements for the Vera Rubin phase, the expansion has an operational update. If the next line is only another capacity announcement, count the megawatts that are actually installed and the GPUs that are running. The figure missing is both.